puffs_vnops.c revision 1.183 1 /* $NetBSD: puffs_vnops.c,v 1.183 2014/08/16 16:19:41 manu Exp $ */
2
3 /*
4 * Copyright (c) 2005, 2006, 2007 Antti Kantee. All Rights Reserved.
5 *
6 * Development of this software was supported by the
7 * Google Summer of Code program and the Ulla Tuominen Foundation.
8 * The Google SoC project was mentored by Bill Studenmund.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
20 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: puffs_vnops.c,v 1.183 2014/08/16 16:19:41 manu Exp $");
34
35 #include <sys/param.h>
36 #include <sys/buf.h>
37 #include <sys/lockf.h>
38 #include <sys/malloc.h>
39 #include <sys/mount.h>
40 #include <sys/namei.h>
41 #include <sys/vnode.h>
42 #include <sys/proc.h>
43 #include <sys/kernel.h> /* For hz, hardclock_ticks */
44
45 #include <uvm/uvm.h>
46
47 #include <fs/puffs/puffs_msgif.h>
48 #include <fs/puffs/puffs_sys.h>
49
50 #include <miscfs/fifofs/fifo.h>
51 #include <miscfs/genfs/genfs.h>
52 #include <miscfs/specfs/specdev.h>
53
54 int puffs_vnop_lookup(void *);
55 int puffs_vnop_create(void *);
56 int puffs_vnop_access(void *);
57 int puffs_vnop_mknod(void *);
58 int puffs_vnop_open(void *);
59 int puffs_vnop_close(void *);
60 int puffs_vnop_getattr(void *);
61 int puffs_vnop_setattr(void *);
62 int puffs_vnop_reclaim(void *);
63 int puffs_vnop_readdir(void *);
64 int puffs_vnop_poll(void *);
65 int puffs_vnop_fsync(void *);
66 int puffs_vnop_seek(void *);
67 int puffs_vnop_remove(void *);
68 int puffs_vnop_mkdir(void *);
69 int puffs_vnop_rmdir(void *);
70 int puffs_vnop_link(void *);
71 int puffs_vnop_readlink(void *);
72 int puffs_vnop_symlink(void *);
73 int puffs_vnop_rename(void *);
74 int puffs_vnop_read(void *);
75 int puffs_vnop_write(void *);
76 int puffs_vnop_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_fallocate_desc, genfs_eopnotsupp }, /* fallocate */
117 { &vop_fdiscard_desc, genfs_eopnotsupp }, /* fdiscard */
118 { &vop_fsync_desc, puffs_vnop_fsync }, /* REAL fsync */
119 { &vop_seek_desc, puffs_vnop_checkop }, /* seek */
120 { &vop_remove_desc, puffs_vnop_checkop }, /* remove */
121 { &vop_link_desc, puffs_vnop_checkop }, /* link */
122 { &vop_rename_desc, puffs_vnop_checkop }, /* rename */
123 { &vop_mkdir_desc, puffs_vnop_checkop }, /* mkdir */
124 { &vop_rmdir_desc, puffs_vnop_checkop }, /* rmdir */
125 { &vop_symlink_desc, puffs_vnop_checkop }, /* symlink */
126 { &vop_readdir_desc, puffs_vnop_checkop }, /* readdir */
127 { &vop_readlink_desc, puffs_vnop_checkop }, /* readlink */
128 { &vop_getpages_desc, puffs_vnop_checkop }, /* getpages */
129 { &vop_putpages_desc, genfs_putpages }, /* REAL putpages */
130 { &vop_pathconf_desc, puffs_vnop_checkop }, /* pathconf */
131 { &vop_advlock_desc, puffs_vnop_advlock }, /* advlock */
132 { &vop_strategy_desc, puffs_vnop_strategy }, /* REAL strategy */
133 { &vop_revoke_desc, genfs_revoke }, /* REAL revoke */
134 { &vop_abortop_desc, puffs_vnop_abortop }, /* REAL abortop */
135 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
136 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
137 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
138 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
139 { &vop_bmap_desc, puffs_vnop_bmap }, /* REAL bmap */
140 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
141 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
142 { &vop_bwrite_desc, genfs_nullop }, /* REAL bwrite */
143 { &vop_mmap_desc, puffs_vnop_mmap }, /* REAL mmap */
144 { &vop_poll_desc, puffs_vnop_poll }, /* REAL poll */
145 { &vop_getextattr_desc, puffs_vnop_getextattr }, /* getextattr */
146 { &vop_setextattr_desc, puffs_vnop_setextattr }, /* setextattr */
147 { &vop_listextattr_desc, puffs_vnop_listextattr }, /* listextattr */
148 { &vop_deleteextattr_desc, puffs_vnop_deleteextattr },/* deleteextattr */
149 #if 0
150 { &vop_openextattr_desc, puffs_vnop_checkop }, /* openextattr */
151 { &vop_closeextattr_desc, puffs_vnop_checkop }, /* closeextattr */
152 #endif
153 { &vop_kqfilter_desc, genfs_eopnotsupp }, /* kqfilter XXX */
154 { NULL, NULL }
155 };
156 const struct vnodeopv_desc puffs_vnodeop_opv_desc =
157 { &puffs_vnodeop_p, puffs_vnodeop_entries };
158
159
160 int (**puffs_specop_p)(void *);
161 const struct vnodeopv_entry_desc puffs_specop_entries[] = {
162 { &vop_default_desc, vn_default_error },
163 { &vop_lookup_desc, spec_lookup }, /* lookup, ENOTDIR */
164 { &vop_create_desc, spec_create }, /* genfs_badop */
165 { &vop_mknod_desc, spec_mknod }, /* genfs_badop */
166 { &vop_open_desc, spec_open }, /* spec_open */
167 { &vop_close_desc, spec_close }, /* spec_close */
168 { &vop_access_desc, puffs_vnop_checkop }, /* access */
169 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
170 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
171 { &vop_read_desc, puffs_vnop_spec_read }, /* update, read */
172 { &vop_write_desc, puffs_vnop_spec_write }, /* update, write */
173 { &vop_fallocate_desc, spec_fallocate }, /* fallocate */
174 { &vop_fdiscard_desc, spec_fdiscard }, /* fdiscard */
175 { &vop_ioctl_desc, spec_ioctl }, /* spec_ioctl */
176 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
177 { &vop_poll_desc, spec_poll }, /* spec_poll */
178 { &vop_kqfilter_desc, spec_kqfilter }, /* spec_kqfilter */
179 { &vop_revoke_desc, spec_revoke }, /* genfs_revoke */
180 { &vop_mmap_desc, spec_mmap }, /* spec_mmap */
181 { &vop_fsync_desc, spec_fsync }, /* vflushbuf */
182 { &vop_seek_desc, spec_seek }, /* genfs_nullop */
183 { &vop_remove_desc, spec_remove }, /* genfs_badop */
184 { &vop_link_desc, spec_link }, /* genfs_badop */
185 { &vop_rename_desc, spec_rename }, /* genfs_badop */
186 { &vop_mkdir_desc, spec_mkdir }, /* genfs_badop */
187 { &vop_rmdir_desc, spec_rmdir }, /* genfs_badop */
188 { &vop_symlink_desc, spec_symlink }, /* genfs_badop */
189 { &vop_readdir_desc, spec_readdir }, /* genfs_badop */
190 { &vop_readlink_desc, spec_readlink }, /* genfs_badop */
191 { &vop_abortop_desc, spec_abortop }, /* genfs_badop */
192 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
193 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
194 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
195 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
196 { &vop_bmap_desc, spec_bmap }, /* dummy */
197 { &vop_strategy_desc, spec_strategy }, /* dev strategy */
198 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
199 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
200 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
201 { &vop_advlock_desc, spec_advlock }, /* lf_advlock */
202 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
203 { &vop_getpages_desc, spec_getpages }, /* genfs_getpages */
204 { &vop_putpages_desc, spec_putpages }, /* genfs_putpages */
205 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
206 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
207 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
208 { &vop_deleteextattr_desc, puffs_vnop_checkop },/* deleteextattr */
209 #if 0
210 { &vop_openextattr_desc, _openextattr }, /* openextattr */
211 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
212 #endif
213 { NULL, NULL }
214 };
215 const struct vnodeopv_desc puffs_specop_opv_desc =
216 { &puffs_specop_p, puffs_specop_entries };
217
218
219 int (**puffs_fifoop_p)(void *);
220 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = {
221 { &vop_default_desc, vn_default_error },
222 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup, ENOTDIR */
223 { &vop_create_desc, vn_fifo_bypass }, /* genfs_badop */
224 { &vop_mknod_desc, vn_fifo_bypass }, /* genfs_badop */
225 { &vop_open_desc, vn_fifo_bypass }, /* open */
226 { &vop_close_desc, vn_fifo_bypass }, /* close */
227 { &vop_access_desc, puffs_vnop_checkop }, /* access */
228 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
229 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
230 { &vop_read_desc, puffs_vnop_fifo_read }, /* read, update */
231 { &vop_write_desc, puffs_vnop_fifo_write }, /* write, update */
232 { &vop_fallocate_desc, vn_fifo_bypass }, /* fallocate */
233 { &vop_fdiscard_desc, vn_fifo_bypass }, /* fdiscard */
234 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
235 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
236 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
237 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
238 { &vop_revoke_desc, vn_fifo_bypass }, /* genfs_revoke */
239 { &vop_mmap_desc, vn_fifo_bypass }, /* genfs_badop */
240 { &vop_fsync_desc, vn_fifo_bypass }, /* genfs_nullop*/
241 { &vop_seek_desc, vn_fifo_bypass }, /* genfs_badop */
242 { &vop_remove_desc, vn_fifo_bypass }, /* genfs_badop */
243 { &vop_link_desc, vn_fifo_bypass }, /* genfs_badop */
244 { &vop_rename_desc, vn_fifo_bypass }, /* genfs_badop */
245 { &vop_mkdir_desc, vn_fifo_bypass }, /* genfs_badop */
246 { &vop_rmdir_desc, vn_fifo_bypass }, /* genfs_badop */
247 { &vop_symlink_desc, vn_fifo_bypass }, /* genfs_badop */
248 { &vop_readdir_desc, vn_fifo_bypass }, /* genfs_badop */
249 { &vop_readlink_desc, vn_fifo_bypass }, /* genfs_badop */
250 { &vop_abortop_desc, vn_fifo_bypass }, /* genfs_badop */
251 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
252 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
253 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
254 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
255 { &vop_bmap_desc, vn_fifo_bypass }, /* dummy */
256 { &vop_strategy_desc, vn_fifo_bypass }, /* genfs_badop */
257 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
258 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
259 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
260 { &vop_advlock_desc, vn_fifo_bypass }, /* genfs_einval */
261 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
262 { &vop_putpages_desc, vn_fifo_bypass }, /* genfs_null_putpages*/
263 #if 0
264 { &vop_openextattr_desc, _openextattr }, /* openextattr */
265 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
266 #endif
267 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
268 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
269 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
270 { &vop_deleteextattr_desc, puffs_vnop_checkop }, /* deleteextattr */
271 { NULL, NULL }
272 };
273 const struct vnodeopv_desc puffs_fifoop_opv_desc =
274 { &puffs_fifoop_p, puffs_fifoop_entries };
275
276
277 /* "real" vnode operations */
278 int (**puffs_msgop_p)(void *);
279 const struct vnodeopv_entry_desc puffs_msgop_entries[] = {
280 { &vop_default_desc, vn_default_error },
281 { &vop_create_desc, puffs_vnop_create }, /* create */
282 { &vop_mknod_desc, puffs_vnop_mknod }, /* mknod */
283 { &vop_open_desc, puffs_vnop_open }, /* open */
284 { &vop_close_desc, puffs_vnop_close }, /* close */
285 { &vop_access_desc, puffs_vnop_access }, /* access */
286 { &vop_getattr_desc, puffs_vnop_getattr }, /* getattr */
287 { &vop_setattr_desc, puffs_vnop_setattr }, /* setattr */
288 { &vop_read_desc, puffs_vnop_read }, /* read */
289 { &vop_write_desc, puffs_vnop_write }, /* write */
290 { &vop_seek_desc, puffs_vnop_seek }, /* seek */
291 { &vop_remove_desc, puffs_vnop_remove }, /* remove */
292 { &vop_link_desc, puffs_vnop_link }, /* link */
293 { &vop_rename_desc, puffs_vnop_rename }, /* rename */
294 { &vop_mkdir_desc, puffs_vnop_mkdir }, /* mkdir */
295 { &vop_rmdir_desc, puffs_vnop_rmdir }, /* rmdir */
296 { &vop_symlink_desc, puffs_vnop_symlink }, /* symlink */
297 { &vop_readdir_desc, puffs_vnop_readdir }, /* readdir */
298 { &vop_readlink_desc, puffs_vnop_readlink }, /* readlink */
299 { &vop_print_desc, puffs_vnop_print }, /* print */
300 { &vop_islocked_desc, puffs_vnop_islocked }, /* islocked */
301 { &vop_pathconf_desc, puffs_vnop_pathconf }, /* pathconf */
302 { &vop_getpages_desc, puffs_vnop_getpages }, /* getpages */
303 { NULL, NULL }
304 };
305 const struct vnodeopv_desc puffs_msgop_opv_desc =
306 { &puffs_msgop_p, puffs_msgop_entries };
307
308 /*
309 * for dosetattr / update_va
310 */
311 #define SETATTR_CHSIZE 0x01
312 #define SETATTR_ASYNC 0x02
313
314 #define ERROUT(err) \
315 do { \
316 error = err; \
317 goto out; \
318 } while (/*CONSTCOND*/0)
319
320 /*
321 * This is a generic vnode operation handler. It checks if the necessary
322 * operations for the called vnode operation are implemented by userspace
323 * and either returns a dummy return value or proceeds to call the real
324 * vnode operation from puffs_msgop_v.
325 *
326 * XXX: this should described elsewhere and autogenerated, the complexity
327 * of the vnode operations vectors and their interrelationships is also
328 * getting a bit out of hand. Another problem is that we need this same
329 * information in the fs server code, so keeping the two in sync manually
330 * is not a viable (long term) plan.
331 */
332
333 /* not supported, handle locking protocol */
334 #define CHECKOP_NOTSUPP(op) \
335 case VOP_##op##_DESCOFFSET: \
336 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
337 return genfs_eopnotsupp(v); \
338 break
339
340 /* always succeed, no locking */
341 #define CHECKOP_SUCCESS(op) \
342 case VOP_##op##_DESCOFFSET: \
343 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
344 return 0; \
345 break
346
347 int
348 puffs_vnop_checkop(void *v)
349 {
350 struct vop_generic_args /* {
351 struct vnodeop_desc *a_desc;
352 spooky mystery contents;
353 } */ *ap = v;
354 struct vnodeop_desc *desc = ap->a_desc;
355 struct puffs_mount *pmp;
356 struct vnode *vp;
357 int offset, rv;
358
359 offset = ap->a_desc->vdesc_vp_offsets[0];
360 #ifdef DIAGNOSTIC
361 if (offset == VDESC_NO_OFFSET)
362 panic("puffs_checkop: no vnode, why did you call me?");
363 #endif
364 vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
365 pmp = MPTOPUFFSMP(vp->v_mount);
366
367 DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n",
368 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp));
369
370 if (!ALLOPS(pmp)) {
371 switch (desc->vdesc_offset) {
372 CHECKOP_NOTSUPP(CREATE);
373 CHECKOP_NOTSUPP(MKNOD);
374 CHECKOP_NOTSUPP(GETATTR);
375 CHECKOP_NOTSUPP(SETATTR);
376 CHECKOP_NOTSUPP(READ);
377 CHECKOP_NOTSUPP(WRITE);
378 CHECKOP_NOTSUPP(FCNTL);
379 CHECKOP_NOTSUPP(IOCTL);
380 CHECKOP_NOTSUPP(REMOVE);
381 CHECKOP_NOTSUPP(LINK);
382 CHECKOP_NOTSUPP(RENAME);
383 CHECKOP_NOTSUPP(MKDIR);
384 CHECKOP_NOTSUPP(RMDIR);
385 CHECKOP_NOTSUPP(SYMLINK);
386 CHECKOP_NOTSUPP(READDIR);
387 CHECKOP_NOTSUPP(READLINK);
388 CHECKOP_NOTSUPP(PRINT);
389 CHECKOP_NOTSUPP(PATHCONF);
390 CHECKOP_NOTSUPP(GETEXTATTR);
391 CHECKOP_NOTSUPP(SETEXTATTR);
392 CHECKOP_NOTSUPP(LISTEXTATTR);
393 CHECKOP_NOTSUPP(DELETEEXTATTR);
394
395 CHECKOP_SUCCESS(ACCESS);
396 CHECKOP_SUCCESS(CLOSE);
397 CHECKOP_SUCCESS(SEEK);
398
399 case VOP_GETPAGES_DESCOFFSET:
400 if (!EXISTSOP(pmp, READ))
401 return genfs_eopnotsupp(v);
402 break;
403
404 default:
405 panic("puffs_checkop: unhandled vnop %d",
406 desc->vdesc_offset);
407 }
408 }
409
410 rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v);
411
412 DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n",
413 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv));
414
415 return rv;
416 }
417
418 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
419 struct componentname *);
420 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
421 struct componentname *);
422 static void callinactive(struct puffs_mount *, puffs_cookie_t, int);
423 static void callreclaim(struct puffs_mount *, puffs_cookie_t, int);
424 static int flushvncache(struct vnode *, off_t, off_t, bool);
425 static void update_va(struct vnode *, struct vattr *, struct vattr *,
426 struct timespec *, struct timespec *, int);
427 static void update_parent(struct vnode *, struct vnode *);
428
429
430 #define PUFFS_ABORT_LOOKUP 1
431 #define PUFFS_ABORT_CREATE 2
432 #define PUFFS_ABORT_MKNOD 3
433 #define PUFFS_ABORT_MKDIR 4
434 #define PUFFS_ABORT_SYMLINK 5
435
436 /*
437 * Press the pani^Wabort button! Kernel resource allocation failed.
438 */
439 static void
440 puffs_abortbutton(struct puffs_mount *pmp, int what,
441 puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp)
442 {
443
444 switch (what) {
445 case PUFFS_ABORT_CREATE:
446 case PUFFS_ABORT_MKNOD:
447 case PUFFS_ABORT_SYMLINK:
448 callremove(pmp, dck, ck, cnp);
449 break;
450 case PUFFS_ABORT_MKDIR:
451 callrmdir(pmp, dck, ck, cnp);
452 break;
453 }
454
455 callinactive(pmp, ck, 0);
456 callreclaim(pmp, ck, 1);
457 }
458
459 /*
460 * Begin vnode operations.
461 *
462 * A word from the keymaster about locks: generally we don't want
463 * to use the vnode locks at all: it creates an ugly dependency between
464 * the userlandia file server and the kernel. But we'll play along with
465 * the kernel vnode locks for now. However, even currently we attempt
466 * to release locks as early as possible. This is possible for some
467 * operations which a) don't need a locked vnode after the userspace op
468 * and b) return with the vnode unlocked. Theoretically we could
469 * unlock-do op-lock for others and order the graph in userspace, but I
470 * don't want to think of the consequences for the time being.
471 */
472
473 #define TTL_TO_TIMEOUT(ts) \
474 (hardclock_ticks + (ts->tv_sec * hz) + (ts->tv_nsec * hz / 1000000000))
475 #define TTL_VALID(ts) \
476 ((ts != NULL) && !((ts->tv_sec == 0) && (ts->tv_nsec == 0)))
477 #define TIMED_OUT(expire) \
478 ((int)((unsigned int)hardclock_ticks - (unsigned int)expire) > 0)
479 int
480 puffs_vnop_lookup(void *v)
481 {
482 struct vop_lookup_v2_args /* {
483 const struct vnodeop_desc *a_desc;
484 struct vnode *a_dvp;
485 struct vnode **a_vpp;
486 struct componentname *a_cnp;
487 } */ *ap = v;
488 PUFFS_MSG_VARS(vn, lookup);
489 struct puffs_mount *pmp;
490 struct componentname *cnp;
491 struct vnode *vp, *dvp, *cvp;
492 struct puffs_node *dpn, *cpn;
493 int isdot;
494 int error;
495
496 pmp = MPTOPUFFSMP(ap->a_dvp->v_mount);
497 cnp = ap->a_cnp;
498 dvp = ap->a_dvp;
499 cvp = NULL;
500 cpn = NULL;
501 *ap->a_vpp = NULL;
502
503 /* r/o fs? we check create later to handle EEXIST */
504 if ((cnp->cn_flags & ISLASTCN)
505 && (dvp->v_mount->mnt_flag & MNT_RDONLY)
506 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
507 return EROFS;
508
509 isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.';
510
511 DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n",
512 cnp->cn_nameptr, dvp, cnp->cn_nameiop));
513
514 /*
515 * If dotdot cache is enabled, add reference to .. and return.
516 */
517 if (PUFFS_USE_DOTDOTCACHE(pmp) && (cnp->cn_flags & ISDOTDOT)) {
518 vp = VPTOPP(ap->a_dvp)->pn_parent;
519 vref(vp);
520
521 *ap->a_vpp = vp;
522 return 0;
523 }
524
525 /*
526 * Check if someone fed it into the cache
527 */
528 if (!isdot && PUFFS_USE_NAMECACHE(pmp)) {
529 int found, iswhiteout;
530
531 found = cache_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
532 cnp->cn_nameiop, cnp->cn_flags,
533 &iswhiteout, ap->a_vpp);
534 if (iswhiteout) {
535 cnp->cn_flags |= ISWHITEOUT;
536 }
537
538 if (found && *ap->a_vpp != NULLVP && PUFFS_USE_FS_TTL(pmp)) {
539 cvp = *ap->a_vpp;
540 cpn = VPTOPP(cvp);
541
542 if (TIMED_OUT(cpn->pn_cn_timeout)) {
543 cache_purge(cvp);
544 /*
545 * cached vnode (cvp) is still referenced
546 * so that we can reuse it upon a new
547 * successful lookup.
548 */
549 *ap->a_vpp = NULL;
550 found = 0;
551 }
552 }
553
554 /*
555 * Do not use negative caching, since the filesystem
556 * provides no TTL for it.
557 */
558 if (found && *ap->a_vpp == NULLVP && PUFFS_USE_FS_TTL(pmp))
559 found = 0;
560
561 if (found) {
562 return *ap->a_vpp == NULLVP ? ENOENT : 0;
563 }
564
565 /*
566 * This is what would have been left in ERROR before
567 * the rearrangement of cache_lookup(). What with all
568 * the macros, I am not sure if this is a dead value
569 * below or not.
570 */
571 error = -1;
572 }
573
574 if (isdot) {
575 /* deal with rename lookup semantics */
576 if (cnp->cn_nameiop == RENAME && (cnp->cn_flags & ISLASTCN))
577 return EISDIR;
578
579 vp = ap->a_dvp;
580 vref(vp);
581 *ap->a_vpp = vp;
582 return 0;
583 }
584
585 if (cvp != NULL) {
586 if (vn_lock(cvp, LK_EXCLUSIVE) != 0) {
587 vrele(cvp);
588 cvp = NULL;
589 } else
590 mutex_enter(&cpn->pn_sizemtx);
591 }
592
593 PUFFS_MSG_ALLOC(vn, lookup);
594 puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred,
595 cnp, PUFFS_USE_FULLPNBUF(pmp));
596
597 if (cnp->cn_flags & ISDOTDOT)
598 VOP_UNLOCK(dvp);
599
600 puffs_msg_setinfo(park_lookup, PUFFSOP_VN,
601 PUFFS_VN_LOOKUP, VPTOPNC(dvp));
602 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error);
603 DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error));
604
605 /*
606 * In case of error, there is no new vnode to play with, so be
607 * happy with the NULL value given to vpp in the beginning.
608 * Also, check if this really was an error or the target was not
609 * present. Either treat it as a non-error for CREATE/RENAME or
610 * enter the component into the negative name cache (if desired).
611 */
612 if (error) {
613 error = checkerr(pmp, error, __func__);
614 if (error == ENOENT) {
615 /* don't allow to create files on r/o fs */
616 if ((dvp->v_mount->mnt_flag & MNT_RDONLY)
617 && cnp->cn_nameiop == CREATE) {
618 error = EROFS;
619
620 /* adjust values if we are creating */
621 } else if ((cnp->cn_flags & ISLASTCN)
622 && (cnp->cn_nameiop == CREATE
623 || cnp->cn_nameiop == RENAME)) {
624 error = EJUSTRETURN;
625
626 /* save negative cache entry */
627 } else {
628 if (PUFFS_USE_NAMECACHE(pmp) &&
629 !PUFFS_USE_FS_TTL(pmp))
630 cache_enter(dvp, NULL, cnp->cn_nameptr,
631 cnp->cn_namelen, cnp->cn_flags);
632 }
633 }
634 goto out;
635 }
636
637 /*
638 * Check that we don't get our parent node back, that would cause
639 * a pretty obvious deadlock.
640 */
641 dpn = dvp->v_data;
642 if (lookup_msg->pvnr_newnode == dpn->pn_cookie) {
643 puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL,
644 "lookup produced parent cookie", lookup_msg->pvnr_newnode);
645 error = EPROTO;
646 goto out;
647 }
648
649 /*
650 * Check if we looked up the cached vnode
651 */
652 vp = NULL;
653 if (cvp && (VPTOPP(cvp)->pn_cookie == lookup_msg->pvnr_newnode)) {
654 int grace;
655
656 /*
657 * Bump grace time of this node so that it does not get
658 * reclaimed too fast. We try to increase a bit more the
659 * lifetime of busiest * nodes - with some limits.
660 */
661 grace = 10 * puffs_sopreq_expire_timeout;
662 cpn->pn_cn_grace = hardclock_ticks + grace;
663 vp = cvp;
664 }
665
666 /*
667 * No cached vnode available, or the cached vnode does not
668 * match the userland cookie anymore: is the node known?
669 */
670 if (vp == NULL) {
671 error = puffs_cookie2vnode(pmp, lookup_msg->pvnr_newnode,
672 1, 1, &vp);
673 }
674
675 if (error == PUFFS_NOSUCHCOOKIE) {
676 error = puffs_getvnode(dvp->v_mount,
677 lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype,
678 lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp);
679 if (error) {
680 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP,
681 VPTOPNC(dvp), lookup_msg->pvnr_newnode,
682 ap->a_cnp);
683 goto out;
684 }
685
686 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
687 } else if (error) {
688 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP, VPTOPNC(dvp),
689 lookup_msg->pvnr_newnode, ap->a_cnp);
690 goto out;
691 }
692
693 /*
694 * Update cache and TTL
695 */
696 if (PUFFS_USE_FS_TTL(pmp)) {
697 struct timespec *va_ttl = &lookup_msg->pvnr_va_ttl;
698 struct timespec *cn_ttl = &lookup_msg->pvnr_cn_ttl;
699 update_va(vp, NULL, &lookup_msg->pvnr_va,
700 va_ttl, cn_ttl, SETATTR_CHSIZE);
701 }
702
703 KASSERT(lookup_msg->pvnr_newnode == VPTOPP(vp)->pn_cookie);
704 *ap->a_vpp = vp;
705
706 if (PUFFS_USE_NAMECACHE(pmp))
707 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
708 cnp->cn_flags);
709
710 /* XXX */
711 if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0)
712 cnp->cn_flags &= ~REQUIREDIR;
713 if (lookup_msg->pvnr_cn.pkcn_consume)
714 cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume,
715 strlen(cnp->cn_nameptr) - cnp->cn_namelen);
716
717 VPTOPP(vp)->pn_nlookup++;
718
719 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
720 (VPTOPP(vp)->pn_parent != dvp))
721 update_parent(vp, dvp);
722
723 out:
724 if (cvp != NULL) {
725 mutex_exit(&cpn->pn_sizemtx);
726
727 if (error || (cvp != vp))
728 vput(cvp);
729 }
730 if (error == 0)
731 VOP_UNLOCK(*ap->a_vpp);
732
733 if (cnp->cn_flags & ISDOTDOT)
734 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
735
736 DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp));
737 PUFFS_MSG_RELEASE(lookup);
738 return error;
739 }
740
741 #define REFPN_AND_UNLOCKVP(a, b) \
742 do { \
743 mutex_enter(&b->pn_mtx); \
744 puffs_referencenode(b); \
745 mutex_exit(&b->pn_mtx); \
746 VOP_UNLOCK(a); \
747 } while (/*CONSTCOND*/0)
748
749 #define REFPN(b) \
750 do { \
751 mutex_enter(&b->pn_mtx); \
752 puffs_referencenode(b); \
753 mutex_exit(&b->pn_mtx); \
754 } while (/*CONSTCOND*/0)
755
756 #define RELEPN_AND_VP(a, b) \
757 do { \
758 puffs_releasenode(b); \
759 vrele(a); \
760 } while (/*CONSTCOND*/0)
761
762 int
763 puffs_vnop_create(void *v)
764 {
765 struct vop_create_v3_args /* {
766 const struct vnodeop_desc *a_desc;
767 struct vnode *a_dvp;
768 struct vnode **a_vpp;
769 struct componentname *a_cnp;
770 struct vattr *a_vap;
771 } */ *ap = v;
772 PUFFS_MSG_VARS(vn, create);
773 struct vnode *dvp = ap->a_dvp;
774 struct puffs_node *dpn = VPTOPP(dvp);
775 struct componentname *cnp = ap->a_cnp;
776 struct mount *mp = dvp->v_mount;
777 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
778 int error;
779
780 DPRINTF(("puffs_create: dvp %p, cnp: %s\n",
781 dvp, ap->a_cnp->cn_nameptr));
782
783 PUFFS_MSG_ALLOC(vn, create);
784 puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred,
785 cnp, PUFFS_USE_FULLPNBUF(pmp));
786 create_msg->pvnr_va = *ap->a_vap;
787 puffs_msg_setinfo(park_create, PUFFSOP_VN,
788 PUFFS_VN_CREATE, VPTOPNC(dvp));
789 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error);
790
791 error = checkerr(pmp, error, __func__);
792 if (error)
793 goto out;
794
795 error = puffs_newnode(mp, dvp, ap->a_vpp,
796 create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0);
797 if (error) {
798 puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie,
799 create_msg->pvnr_newnode, cnp);
800 goto out;
801 }
802
803 if (PUFFS_USE_FS_TTL(pmp)) {
804 struct timespec *va_ttl = &create_msg->pvnr_va_ttl;
805 struct timespec *cn_ttl = &create_msg->pvnr_cn_ttl;
806 struct vattr *rvap = &create_msg->pvnr_va;
807
808 update_va(*ap->a_vpp, NULL, rvap,
809 va_ttl, cn_ttl, SETATTR_CHSIZE);
810 }
811
812 VPTOPP(*ap->a_vpp)->pn_nlookup++;
813
814 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
815 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
816 update_parent(*ap->a_vpp, dvp);
817
818 out:
819 DPRINTF(("puffs_create: return %d\n", error));
820 PUFFS_MSG_RELEASE(create);
821 return error;
822 }
823
824 int
825 puffs_vnop_mknod(void *v)
826 {
827 struct vop_mknod_v3_args /* {
828 const struct vnodeop_desc *a_desc;
829 struct vnode *a_dvp;
830 struct vnode **a_vpp;
831 struct componentname *a_cnp;
832 struct vattr *a_vap;
833 } */ *ap = v;
834 PUFFS_MSG_VARS(vn, mknod);
835 struct vnode *dvp = ap->a_dvp;
836 struct puffs_node *dpn = VPTOPP(dvp);
837 struct componentname *cnp = ap->a_cnp;
838 struct mount *mp = dvp->v_mount;
839 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
840 int error;
841
842 PUFFS_MSG_ALLOC(vn, mknod);
843 puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred,
844 cnp, PUFFS_USE_FULLPNBUF(pmp));
845 mknod_msg->pvnr_va = *ap->a_vap;
846 puffs_msg_setinfo(park_mknod, PUFFSOP_VN,
847 PUFFS_VN_MKNOD, VPTOPNC(dvp));
848
849 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error);
850
851 error = checkerr(pmp, error, __func__);
852 if (error)
853 goto out;
854
855 error = puffs_newnode(mp, dvp, ap->a_vpp,
856 mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type,
857 ap->a_vap->va_rdev);
858 if (error) {
859 puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie,
860 mknod_msg->pvnr_newnode, cnp);
861 goto out;
862 }
863
864 if (PUFFS_USE_FS_TTL(pmp)) {
865 struct timespec *va_ttl = &mknod_msg->pvnr_va_ttl;
866 struct timespec *cn_ttl = &mknod_msg->pvnr_cn_ttl;
867 struct vattr *rvap = &mknod_msg->pvnr_va;
868
869 update_va(*ap->a_vpp, NULL, rvap,
870 va_ttl, cn_ttl, SETATTR_CHSIZE);
871 }
872
873 VPTOPP(*ap->a_vpp)->pn_nlookup++;
874
875 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
876 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
877 update_parent(*ap->a_vpp, dvp);
878
879 out:
880 PUFFS_MSG_RELEASE(mknod);
881 return error;
882 }
883
884 int
885 puffs_vnop_open(void *v)
886 {
887 struct vop_open_args /* {
888 const struct vnodeop_desc *a_desc;
889 struct vnode *a_vp;
890 int a_mode;
891 kauth_cred_t a_cred;
892 } */ *ap = v;
893 PUFFS_MSG_VARS(vn, open);
894 struct vnode *vp = ap->a_vp;
895 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
896 struct puffs_node *pn = VPTOPP(vp);
897 int mode = ap->a_mode;
898 int error;
899
900 DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode));
901
902 if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE))
903 ERROUT(EROFS);
904
905 if (!EXISTSOP(pmp, OPEN))
906 ERROUT(0);
907
908 PUFFS_MSG_ALLOC(vn, open);
909 open_msg->pvnr_mode = mode;
910 puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred);
911 puffs_msg_setinfo(park_open, PUFFSOP_VN,
912 PUFFS_VN_OPEN, VPTOPNC(vp));
913
914 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error);
915 error = checkerr(pmp, error, __func__);
916
917 if (open_msg->pvnr_oflags & PUFFS_OPEN_IO_DIRECT) {
918 if (mode & FREAD)
919 pn->pn_stat |= PNODE_RDIRECT;
920 if (mode & FWRITE)
921 pn->pn_stat |= PNODE_WDIRECT;
922 }
923 out:
924 DPRINTF(("puffs_open: returning %d\n", error));
925 PUFFS_MSG_RELEASE(open);
926 return error;
927 }
928
929 int
930 puffs_vnop_close(void *v)
931 {
932 struct vop_close_args /* {
933 const struct vnodeop_desc *a_desc;
934 struct vnode *a_vp;
935 int a_fflag;
936 kauth_cred_t a_cred;
937 } */ *ap = v;
938 PUFFS_MSG_VARS(vn, close);
939 struct vnode *vp = ap->a_vp;
940 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
941
942 PUFFS_MSG_ALLOC(vn, close);
943 puffs_msg_setfaf(park_close);
944 close_msg->pvnr_fflag = ap->a_fflag;
945 puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
946 puffs_msg_setinfo(park_close, PUFFSOP_VN,
947 PUFFS_VN_CLOSE, VPTOPNC(vp));
948
949 puffs_msg_enqueue(pmp, park_close);
950 PUFFS_MSG_RELEASE(close);
951 return 0;
952 }
953
954 int
955 puffs_vnop_access(void *v)
956 {
957 struct vop_access_args /* {
958 const struct vnodeop_desc *a_desc;
959 struct vnode *a_vp;
960 int a_mode;
961 kauth_cred_t a_cred;
962 } */ *ap = v;
963 PUFFS_MSG_VARS(vn, access);
964 struct vnode *vp = ap->a_vp;
965 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
966 int mode = ap->a_mode;
967 int error;
968
969 if (mode & VWRITE) {
970 switch (vp->v_type) {
971 case VDIR:
972 case VLNK:
973 case VREG:
974 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
975 || !EXISTSOP(pmp, WRITE))
976 return EROFS;
977 break;
978 default:
979 break;
980 }
981 }
982
983 if (!EXISTSOP(pmp, ACCESS))
984 return 0;
985
986 PUFFS_MSG_ALLOC(vn, access);
987 access_msg->pvnr_mode = ap->a_mode;
988 puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
989 puffs_msg_setinfo(park_access, PUFFSOP_VN,
990 PUFFS_VN_ACCESS, VPTOPNC(vp));
991
992 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
993 error = checkerr(pmp, error, __func__);
994 PUFFS_MSG_RELEASE(access);
995
996 return error;
997 }
998
999 static void
1000 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap,
1001 struct timespec *va_ttl, struct timespec *cn_ttl, int flags)
1002 {
1003 struct puffs_node *pn = VPTOPP(vp);
1004
1005 if (TTL_VALID(cn_ttl)) {
1006 pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl);
1007 pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace);
1008 }
1009
1010 /*
1011 * Don't listen to the file server regarding special device
1012 * size info, the file server doesn't know anything about them.
1013 */
1014 if (vp->v_type == VBLK || vp->v_type == VCHR)
1015 rvap->va_size = vp->v_size;
1016
1017 /* Ditto for blocksize (ufs comment: this doesn't belong here) */
1018 if (vp->v_type == VBLK)
1019 rvap->va_blocksize = BLKDEV_IOSIZE;
1020 else if (vp->v_type == VCHR)
1021 rvap->va_blocksize = MAXBSIZE;
1022
1023 if (vap != NULL) {
1024 (void) memcpy(vap, rvap, sizeof(struct vattr));
1025 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1026
1027 if (pn->pn_stat & PNODE_METACACHE_ATIME)
1028 vap->va_atime = pn->pn_mc_atime;
1029 if (pn->pn_stat & PNODE_METACACHE_CTIME)
1030 vap->va_ctime = pn->pn_mc_ctime;
1031 if (pn->pn_stat & PNODE_METACACHE_MTIME)
1032 vap->va_mtime = pn->pn_mc_mtime;
1033 if (pn->pn_stat & PNODE_METACACHE_SIZE)
1034 vap->va_size = pn->pn_mc_size;
1035 }
1036
1037 if (!(pn->pn_stat & PNODE_METACACHE_SIZE) && (flags & SETATTR_CHSIZE)) {
1038 if (rvap->va_size != VNOVAL
1039 && vp->v_type != VBLK && vp->v_type != VCHR) {
1040 uvm_vnp_setsize(vp, rvap->va_size);
1041 pn->pn_serversize = rvap->va_size;
1042 }
1043 }
1044
1045 if ((va_ttl != NULL) && TTL_VALID(va_ttl)) {
1046 if (pn->pn_va_cache == NULL)
1047 pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK);
1048
1049 (void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap));
1050
1051 pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl);
1052 }
1053 }
1054
1055 static void
1056 update_parent(struct vnode *vp, struct vnode *dvp)
1057 {
1058 struct puffs_node *pn = VPTOPP(vp);
1059
1060 if (pn->pn_parent != NULL) {
1061 KASSERT(pn->pn_parent != dvp);
1062 vrele(pn->pn_parent);
1063 }
1064
1065 vref(dvp);
1066 pn->pn_parent = dvp;
1067 }
1068
1069 int
1070 puffs_vnop_getattr(void *v)
1071 {
1072 struct vop_getattr_args /* {
1073 const struct vnodeop_desc *a_desc;
1074 struct vnode *a_vp;
1075 struct vattr *a_vap;
1076 kauth_cred_t a_cred;
1077 } */ *ap = v;
1078 PUFFS_MSG_VARS(vn, getattr);
1079 struct vnode *vp = ap->a_vp;
1080 struct mount *mp = vp->v_mount;
1081 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1082 struct vattr *vap, *rvap;
1083 struct puffs_node *pn = VPTOPP(vp);
1084 struct timespec *va_ttl = NULL;
1085 int error = 0;
1086
1087 /*
1088 * A lock is required so that we do not race with
1089 * setattr, write and fsync when changing vp->v_size.
1090 * This is critical, since setting a stall smaler value
1091 * triggers a file truncate in uvm_vnp_setsize(), which
1092 * most of the time means data corruption (a chunk of
1093 * data is replaced by zeroes). This can be removed if
1094 * we decide one day that VOP_GETATTR must operate on
1095 * a locked vnode.
1096 *
1097 * XXX Should be useless now that VOP_GETATTR has been
1098 * fixed to always require a shared lock at least.
1099 */
1100 mutex_enter(&pn->pn_sizemtx);
1101
1102 REFPN(pn);
1103 vap = ap->a_vap;
1104
1105 if (PUFFS_USE_FS_TTL(pmp)) {
1106 if (!TIMED_OUT(pn->pn_va_timeout)) {
1107 update_va(vp, vap, pn->pn_va_cache,
1108 NULL, NULL, SETATTR_CHSIZE);
1109 goto out2;
1110 }
1111 }
1112
1113 PUFFS_MSG_ALLOC(vn, getattr);
1114 vattr_null(&getattr_msg->pvnr_va);
1115 puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
1116 puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
1117 PUFFS_VN_GETATTR, VPTOPNC(vp));
1118
1119 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
1120 error = checkerr(pmp, error, __func__);
1121 if (error)
1122 goto out;
1123
1124 rvap = &getattr_msg->pvnr_va;
1125
1126 if (PUFFS_USE_FS_TTL(pmp))
1127 va_ttl = &getattr_msg->pvnr_va_ttl;
1128
1129 update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE);
1130
1131 out:
1132 PUFFS_MSG_RELEASE(getattr);
1133
1134 out2:
1135 puffs_releasenode(pn);
1136
1137 mutex_exit(&pn->pn_sizemtx);
1138
1139 return error;
1140 }
1141
1142 static int
1143 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
1144 {
1145 PUFFS_MSG_VARS(vn, setattr);
1146 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1147 struct puffs_node *pn = vp->v_data;
1148 int error = 0;
1149
1150 KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx));
1151
1152 if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
1153 (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
1154 || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
1155 || vap->va_mode != (mode_t)VNOVAL))
1156 return EROFS;
1157
1158 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
1159 && vp->v_type == VREG && vap->va_size != VNOVAL)
1160 return EROFS;
1161
1162 /*
1163 * Flush metacache first. If we are called with some explicit
1164 * parameters, treat them as information overriding metacache
1165 * information.
1166 */
1167 if (pn->pn_stat & PNODE_METACACHE_MASK) {
1168 if ((pn->pn_stat & PNODE_METACACHE_ATIME)
1169 && vap->va_atime.tv_sec == VNOVAL)
1170 vap->va_atime = pn->pn_mc_atime;
1171 if ((pn->pn_stat & PNODE_METACACHE_CTIME)
1172 && vap->va_ctime.tv_sec == VNOVAL)
1173 vap->va_ctime = pn->pn_mc_ctime;
1174 if ((pn->pn_stat & PNODE_METACACHE_MTIME)
1175 && vap->va_mtime.tv_sec == VNOVAL)
1176 vap->va_mtime = pn->pn_mc_mtime;
1177 if ((pn->pn_stat & PNODE_METACACHE_SIZE)
1178 && vap->va_size == VNOVAL)
1179 vap->va_size = pn->pn_mc_size;
1180
1181 pn->pn_stat &= ~PNODE_METACACHE_MASK;
1182 }
1183
1184 /*
1185 * Flush attribute cache so that another thread do
1186 * not get a stale value during the operation.
1187 */
1188 if (PUFFS_USE_FS_TTL(pmp))
1189 pn->pn_va_timeout = 0;
1190
1191 PUFFS_MSG_ALLOC(vn, setattr);
1192 (void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
1193 puffs_credcvt(&setattr_msg->pvnr_cred, cred);
1194 puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
1195 PUFFS_VN_SETATTR, VPTOPNC(vp));
1196 if (flags & SETATTR_ASYNC)
1197 puffs_msg_setfaf(park_setattr);
1198
1199 puffs_msg_enqueue(pmp, park_setattr);
1200 if ((flags & SETATTR_ASYNC) == 0)
1201 error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
1202
1203 if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) {
1204 struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl;
1205 struct vattr *rvap = &setattr_msg->pvnr_va;
1206
1207 update_va(vp, NULL, rvap, va_ttl, NULL, flags);
1208 }
1209
1210 PUFFS_MSG_RELEASE(setattr);
1211 if ((flags & SETATTR_ASYNC) == 0) {
1212 error = checkerr(pmp, error, __func__);
1213 if (error)
1214 return error;
1215 } else {
1216 error = 0;
1217 }
1218
1219 if (vap->va_size != VNOVAL) {
1220 pn->pn_serversize = vap->va_size;
1221 if (flags & SETATTR_CHSIZE)
1222 uvm_vnp_setsize(vp, vap->va_size);
1223 }
1224
1225 return 0;
1226 }
1227
1228 int
1229 puffs_vnop_setattr(void *v)
1230 {
1231 struct vop_getattr_args /* {
1232 const struct vnodeop_desc *a_desc;
1233 struct vnode *a_vp;
1234 struct vattr *a_vap;
1235 kauth_cred_t a_cred;
1236 } */ *ap = v;
1237 struct puffs_node *pn = ap->a_vp->v_data;
1238 int error;
1239
1240 mutex_enter(&pn->pn_sizemtx);
1241 error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
1242 mutex_exit(&pn->pn_sizemtx);
1243
1244 return error;
1245 }
1246
1247 static __inline int
1248 doinact(struct puffs_mount *pmp, int iaflag)
1249 {
1250
1251 if (EXISTSOP(pmp, INACTIVE))
1252 if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
1253 if (iaflag || ALLOPS(pmp))
1254 return 1;
1255 else
1256 return 0;
1257 else
1258 return 1;
1259 else
1260 return 0;
1261 }
1262
1263 static void
1264 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
1265 {
1266 PUFFS_MSG_VARS(vn, inactive);
1267
1268 if (doinact(pmp, iaflag)) {
1269 PUFFS_MSG_ALLOC(vn, inactive);
1270 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1271 PUFFS_VN_INACTIVE, ck);
1272 PUFFS_MSG_ENQUEUEWAIT_NOERROR(pmp, park_inactive);
1273 PUFFS_MSG_RELEASE(inactive);
1274 }
1275 }
1276
1277 /* XXX: callinactive can't setback */
1278 int
1279 puffs_vnop_inactive(void *v)
1280 {
1281 struct vop_inactive_args /* {
1282 const struct vnodeop_desc *a_desc;
1283 struct vnode *a_vp;
1284 } */ *ap = v;
1285 PUFFS_MSG_VARS(vn, inactive);
1286 struct vnode *vp = ap->a_vp;
1287 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1288 struct puffs_node *pnode;
1289 bool recycle = false;
1290
1291 pnode = vp->v_data;
1292 mutex_enter(&pnode->pn_sizemtx);
1293
1294 if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1295 flushvncache(vp, 0, 0, false);
1296 PUFFS_MSG_ALLOC(vn, inactive);
1297 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1298 PUFFS_VN_INACTIVE, VPTOPNC(vp));
1299 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_inactive, vp->v_data,
1300 NULL);
1301 PUFFS_MSG_RELEASE(inactive);
1302 }
1303 pnode->pn_stat &= ~PNODE_DOINACT;
1304
1305 /*
1306 * file server thinks it's gone? then don't be afraid care,
1307 * node's life was already all it would ever be
1308 */
1309 if (pnode->pn_stat & PNODE_NOREFS) {
1310 pnode->pn_stat |= PNODE_DYING;
1311 recycle = true;
1312 }
1313
1314 /*
1315 * Handle node TTL.
1316 * If grace has already timed out, make it reclaimed.
1317 * Otherwise, we queue its expiration by sop thread, so
1318 * that it does not remain for ages in the freelist,
1319 * holding memory in userspace, while we will have
1320 * to look it up again anyway.
1321 */
1322 if (PUFFS_USE_FS_TTL(pmp) && !(vp->v_vflag & VV_ROOT) && !recycle) {
1323 bool incache = !TIMED_OUT(pnode->pn_cn_timeout);
1324 bool ingrace = !TIMED_OUT(pnode->pn_cn_grace);
1325 bool reclaimqueued = pnode->pn_stat & PNODE_SOPEXP;
1326
1327 if (!incache && !ingrace && !reclaimqueued) {
1328 pnode->pn_stat |= PNODE_DYING;
1329 recycle = true;
1330 }
1331
1332 if (!recycle && !reclaimqueued) {
1333 struct puffs_sopreq *psopr;
1334 int at = MAX(pnode->pn_cn_grace, pnode->pn_cn_timeout);
1335
1336 KASSERT(curlwp != uvm.pagedaemon_lwp);
1337 psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP);
1338 psopr->psopr_ck = VPTOPNC(pnode->pn_vp);
1339 psopr->psopr_sopreq = PUFFS_SOPREQ_EXPIRE;
1340 psopr->psopr_at = at;
1341
1342 mutex_enter(&pmp->pmp_sopmtx);
1343
1344 /*
1345 * If thread has disapeared, just give up. The
1346 * fs is being unmounted and the node will be
1347 * be reclaimed anyway.
1348 *
1349 * Otherwise, we queue the request but do not
1350 * immediatly signal the thread, as the node
1351 * has not been expired yet.
1352 */
1353 if (pmp->pmp_sopthrcount == 0) {
1354 kmem_free(psopr, sizeof(*psopr));
1355 } else {
1356 TAILQ_INSERT_TAIL(&pmp->pmp_sopnodereqs,
1357 psopr, psopr_entries);
1358 pnode->pn_stat |= PNODE_SOPEXP;
1359 }
1360
1361 mutex_exit(&pmp->pmp_sopmtx);
1362 }
1363 }
1364
1365 *ap->a_recycle = recycle;
1366
1367 mutex_exit(&pnode->pn_sizemtx);
1368 VOP_UNLOCK(vp);
1369
1370 return 0;
1371 }
1372
1373 static void
1374 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck, int nlookup)
1375 {
1376 PUFFS_MSG_VARS(vn, reclaim);
1377
1378 if (!EXISTSOP(pmp, RECLAIM))
1379 return;
1380
1381 PUFFS_MSG_ALLOC(vn, reclaim);
1382 reclaim_msg->pvnr_nlookup = nlookup;
1383 puffs_msg_setfaf(park_reclaim);
1384 puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck);
1385
1386 puffs_msg_enqueue(pmp, park_reclaim);
1387 PUFFS_MSG_RELEASE(reclaim);
1388 return;
1389 }
1390
1391 /*
1392 * always FAF, we don't really care if the server wants to fail to
1393 * reclaim the node or not
1394 */
1395 int
1396 puffs_vnop_reclaim(void *v)
1397 {
1398 struct vop_reclaim_args /* {
1399 const struct vnodeop_desc *a_desc;
1400 struct vnode *a_vp;
1401 } */ *ap = v;
1402 struct vnode *vp = ap->a_vp;
1403 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1404 struct puffs_node *pnode = vp->v_data;
1405 bool notifyserver = true;
1406
1407 /*
1408 * first things first: check if someone is trying to reclaim the
1409 * root vnode. do not allow that to travel to userspace.
1410 * Note that we don't need to take the lock similarly to
1411 * puffs_root(), since there is only one of us.
1412 */
1413 if (vp->v_vflag & VV_ROOT) {
1414 mutex_enter(&pmp->pmp_lock);
1415 KASSERT(pmp->pmp_root != NULL);
1416 pmp->pmp_root = NULL;
1417 mutex_exit(&pmp->pmp_lock);
1418 notifyserver = false;
1419 }
1420
1421 /*
1422 * purge info from kernel before issueing FAF, since we
1423 * don't really know when we'll get around to it after
1424 * that and someone might race us into node creation
1425 */
1426 mutex_enter(&pmp->pmp_lock);
1427 LIST_REMOVE(pnode, pn_hashent);
1428 if (PUFFS_USE_NAMECACHE(pmp))
1429 cache_purge(vp);
1430 mutex_exit(&pmp->pmp_lock);
1431
1432 if (notifyserver) {
1433 int nlookup = VPTOPP(vp)->pn_nlookup;
1434
1435 callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp), nlookup);
1436 }
1437
1438 if (PUFFS_USE_DOTDOTCACHE(pmp)) {
1439 if (__predict_true(VPTOPP(vp)->pn_parent != NULL))
1440 vrele(VPTOPP(vp)->pn_parent);
1441 else
1442 KASSERT(vp->v_vflag & VV_ROOT);
1443 }
1444
1445 puffs_putvnode(vp);
1446 vp->v_data = NULL;
1447
1448 return 0;
1449 }
1450
1451 #define CSIZE sizeof(**ap->a_cookies)
1452 int
1453 puffs_vnop_readdir(void *v)
1454 {
1455 struct vop_readdir_args /* {
1456 const struct vnodeop_desc *a_desc;
1457 struct vnode *a_vp;
1458 struct uio *a_uio;
1459 kauth_cred_t a_cred;
1460 int *a_eofflag;
1461 off_t **a_cookies;
1462 int *a_ncookies;
1463 } */ *ap = v;
1464 PUFFS_MSG_VARS(vn, readdir);
1465 struct vnode *vp = ap->a_vp;
1466 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1467 size_t argsize, tomove, cookiemem, cookiesmax;
1468 struct uio *uio = ap->a_uio;
1469 size_t howmuch, resid;
1470 int error;
1471
1472 /*
1473 * ok, so we need: resid + cookiemem = maxreq
1474 * => resid + cookiesize * (resid/minsize) = maxreq
1475 * => resid + cookiesize/minsize * resid = maxreq
1476 * => (cookiesize/minsize + 1) * resid = maxreq
1477 * => resid = maxreq / (cookiesize/minsize + 1)
1478 *
1479 * Since cookiesize <= minsize and we're not very big on floats,
1480 * we approximate that to be 1. Therefore:
1481 *
1482 * resid = maxreq / 2;
1483 *
1484 * Well, at least we didn't have to use differential equations
1485 * or the Gram-Schmidt process.
1486 *
1487 * (yes, I'm very afraid of this)
1488 */
1489 KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0));
1490
1491 if (ap->a_cookies) {
1492 KASSERT(ap->a_ncookies != NULL);
1493 if (pmp->pmp_args.pa_fhsize == 0)
1494 return EOPNOTSUPP;
1495 resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2;
1496 cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0);
1497 cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */
1498 } else {
1499 resid = PUFFS_TOMOVE(uio->uio_resid, pmp);
1500 cookiesmax = 0;
1501 cookiemem = 0;
1502 }
1503
1504 argsize = sizeof(struct puffs_vnmsg_readdir);
1505 tomove = resid + cookiemem;
1506 puffs_msgmem_alloc(argsize + tomove, &park_readdir,
1507 (void *)&readdir_msg, 1);
1508
1509 puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred);
1510 readdir_msg->pvnr_offset = uio->uio_offset;
1511 readdir_msg->pvnr_resid = resid;
1512 readdir_msg->pvnr_ncookies = cookiesmax;
1513 readdir_msg->pvnr_eofflag = 0;
1514 readdir_msg->pvnr_dentoff = cookiemem;
1515 puffs_msg_setinfo(park_readdir, PUFFSOP_VN,
1516 PUFFS_VN_READDIR, VPTOPNC(vp));
1517 puffs_msg_setdelta(park_readdir, tomove);
1518
1519 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error);
1520 error = checkerr(pmp, error, __func__);
1521 if (error)
1522 goto out;
1523
1524 /* userspace is cheating? */
1525 if (readdir_msg->pvnr_resid > resid) {
1526 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1527 "resid grew", VPTOPNC(vp));
1528 ERROUT(EPROTO);
1529 }
1530 if (readdir_msg->pvnr_ncookies > cookiesmax) {
1531 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1532 "too many cookies", VPTOPNC(vp));
1533 ERROUT(EPROTO);
1534 }
1535
1536 /* check eof */
1537 if (readdir_msg->pvnr_eofflag)
1538 *ap->a_eofflag = 1;
1539
1540 /* bouncy-wouncy with the directory data */
1541 howmuch = resid - readdir_msg->pvnr_resid;
1542
1543 /* force eof if no data was returned (getcwd() needs this) */
1544 if (howmuch == 0) {
1545 *ap->a_eofflag = 1;
1546 goto out;
1547 }
1548
1549 error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio);
1550 if (error)
1551 goto out;
1552
1553 /* provide cookies to caller if so desired */
1554 if (ap->a_cookies) {
1555 KASSERT(curlwp != uvm.pagedaemon_lwp);
1556 *ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE,
1557 M_TEMP, M_WAITOK);
1558 *ap->a_ncookies = readdir_msg->pvnr_ncookies;
1559 memcpy(*ap->a_cookies, readdir_msg->pvnr_data,
1560 *ap->a_ncookies*CSIZE);
1561 }
1562
1563 /* next readdir starts here */
1564 uio->uio_offset = readdir_msg->pvnr_offset;
1565
1566 out:
1567 puffs_msgmem_release(park_readdir);
1568 return error;
1569 }
1570 #undef CSIZE
1571
1572 /*
1573 * poll works by consuming the bitmask in pn_revents. If there are
1574 * events available, poll returns immediately. If not, it issues a
1575 * poll to userspace, selrecords itself and returns with no available
1576 * events. When the file server returns, it executes puffs_parkdone_poll(),
1577 * where available events are added to the bitmask. selnotify() is
1578 * then also executed by that function causing us to enter here again
1579 * and hopefully find the missing bits (unless someone got them first,
1580 * in which case it starts all over again).
1581 */
1582 int
1583 puffs_vnop_poll(void *v)
1584 {
1585 struct vop_poll_args /* {
1586 const struct vnodeop_desc *a_desc;
1587 struct vnode *a_vp;
1588 int a_events;
1589 } */ *ap = v;
1590 PUFFS_MSG_VARS(vn, poll);
1591 struct vnode *vp = ap->a_vp;
1592 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1593 struct puffs_node *pn = vp->v_data;
1594 int events;
1595
1596 if (EXISTSOP(pmp, POLL)) {
1597 mutex_enter(&pn->pn_mtx);
1598 events = pn->pn_revents & ap->a_events;
1599 if (events & ap->a_events) {
1600 pn->pn_revents &= ~ap->a_events;
1601 mutex_exit(&pn->pn_mtx);
1602
1603 return events;
1604 } else {
1605 puffs_referencenode(pn);
1606 mutex_exit(&pn->pn_mtx);
1607
1608 PUFFS_MSG_ALLOC(vn, poll);
1609 poll_msg->pvnr_events = ap->a_events;
1610 puffs_msg_setinfo(park_poll, PUFFSOP_VN,
1611 PUFFS_VN_POLL, VPTOPNC(vp));
1612 puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn);
1613 selrecord(curlwp, &pn->pn_sel);
1614
1615 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_poll,
1616 vp->v_data, NULL);
1617 PUFFS_MSG_RELEASE(poll);
1618
1619 return 0;
1620 }
1621 } else {
1622 return genfs_poll(v);
1623 }
1624 }
1625
1626 static int
1627 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait)
1628 {
1629 struct puffs_node *pn = VPTOPP(vp);
1630 struct vattr va;
1631 int pflags, error;
1632
1633 /* flush out information from our metacache, see vop_setattr */
1634 if (pn->pn_stat & PNODE_METACACHE_MASK
1635 && (pn->pn_stat & PNODE_DYING) == 0) {
1636 vattr_null(&va);
1637 error = dosetattr(vp, &va, FSCRED,
1638 SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC));
1639 if (error)
1640 return error;
1641 }
1642
1643 /*
1644 * flush pages to avoid being overly dirty
1645 */
1646 pflags = PGO_CLEANIT;
1647 if (wait)
1648 pflags |= PGO_SYNCIO;
1649
1650 mutex_enter(vp->v_interlock);
1651 return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags);
1652 }
1653
1654 int
1655 puffs_vnop_fsync(void *v)
1656 {
1657 struct vop_fsync_args /* {
1658 const struct vnodeop_desc *a_desc;
1659 struct vnode *a_vp;
1660 kauth_cred_t a_cred;
1661 int a_flags;
1662 off_t a_offlo;
1663 off_t a_offhi;
1664 } */ *ap = v;
1665 PUFFS_MSG_VARS(vn, fsync);
1666 struct vnode *vp;
1667 struct puffs_node *pn;
1668 struct puffs_mount *pmp;
1669 int error, dofaf;
1670
1671 vp = ap->a_vp;
1672 KASSERT(vp != NULL);
1673 pn = VPTOPP(vp);
1674 KASSERT(pn != NULL);
1675 pmp = MPTOPUFFSMP(vp->v_mount);
1676 if (ap->a_flags & FSYNC_WAIT) {
1677 mutex_enter(&pn->pn_sizemtx);
1678 } else {
1679 if (mutex_tryenter(&pn->pn_sizemtx) == 0)
1680 return EDEADLK;
1681 }
1682
1683 error = flushvncache(vp, ap->a_offlo, ap->a_offhi,
1684 (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT);
1685 if (error)
1686 goto out;
1687
1688 /*
1689 * HELLO! We exit already here if the user server does not
1690 * support fsync OR if we should call fsync for a node which
1691 * has references neither in the kernel or the fs server.
1692 * Otherwise we continue to issue fsync() forward.
1693 */
1694 error = 0;
1695 if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING))
1696 goto out;
1697
1698 dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY;
1699 /*
1700 * We abuse VXLOCK to mean "vnode is going to die", so we issue
1701 * only FAFs for those. Otherwise there's a danger of deadlock,
1702 * since the execution context here might be the user server
1703 * doing some operation on another fs, which in turn caused a
1704 * vnode to be reclaimed from the freelist for this fs.
1705 */
1706 if (dofaf == 0) {
1707 mutex_enter(vp->v_interlock);
1708 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
1709 dofaf = 1;
1710 mutex_exit(vp->v_interlock);
1711 }
1712
1713 PUFFS_MSG_ALLOC(vn, fsync);
1714 if (dofaf)
1715 puffs_msg_setfaf(park_fsync);
1716
1717 puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred);
1718 fsync_msg->pvnr_flags = ap->a_flags;
1719 fsync_msg->pvnr_offlo = ap->a_offlo;
1720 fsync_msg->pvnr_offhi = ap->a_offhi;
1721 puffs_msg_setinfo(park_fsync, PUFFSOP_VN,
1722 PUFFS_VN_FSYNC, VPTOPNC(vp));
1723
1724 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error);
1725 PUFFS_MSG_RELEASE(fsync);
1726
1727 error = checkerr(pmp, error, __func__);
1728
1729 out:
1730 mutex_exit(&pn->pn_sizemtx);
1731 return error;
1732 }
1733
1734 int
1735 puffs_vnop_seek(void *v)
1736 {
1737 struct vop_seek_args /* {
1738 const struct vnodeop_desc *a_desc;
1739 struct vnode *a_vp;
1740 off_t a_oldoff;
1741 off_t a_newoff;
1742 kauth_cred_t a_cred;
1743 } */ *ap = v;
1744 PUFFS_MSG_VARS(vn, seek);
1745 struct vnode *vp = ap->a_vp;
1746 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1747 int error;
1748
1749 PUFFS_MSG_ALLOC(vn, seek);
1750 seek_msg->pvnr_oldoff = ap->a_oldoff;
1751 seek_msg->pvnr_newoff = ap->a_newoff;
1752 puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred);
1753 puffs_msg_setinfo(park_seek, PUFFSOP_VN,
1754 PUFFS_VN_SEEK, VPTOPNC(vp));
1755
1756 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error);
1757 PUFFS_MSG_RELEASE(seek);
1758 return checkerr(pmp, error, __func__);
1759 }
1760
1761 static int
1762 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1763 struct componentname *cnp)
1764 {
1765 PUFFS_MSG_VARS(vn, remove);
1766 int error;
1767
1768 PUFFS_MSG_ALLOC(vn, remove);
1769 remove_msg->pvnr_cookie_targ = ck;
1770 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1771 cnp, PUFFS_USE_FULLPNBUF(pmp));
1772 puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck);
1773
1774 PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error);
1775 PUFFS_MSG_RELEASE(remove);
1776
1777 return checkerr(pmp, error, __func__);
1778 }
1779
1780 /*
1781 * XXX: can't use callremove now because can't catch setbacks with
1782 * it due to lack of a pnode argument.
1783 */
1784 int
1785 puffs_vnop_remove(void *v)
1786 {
1787 struct vop_remove_args /* {
1788 const struct vnodeop_desc *a_desc;
1789 struct vnode *a_dvp;
1790 struct vnode *a_vp;
1791 struct componentname *a_cnp;
1792 } */ *ap = v;
1793 PUFFS_MSG_VARS(vn, remove);
1794 struct vnode *dvp = ap->a_dvp;
1795 struct vnode *vp = ap->a_vp;
1796 struct puffs_node *dpn = VPTOPP(dvp);
1797 struct puffs_node *pn = VPTOPP(vp);
1798 struct componentname *cnp = ap->a_cnp;
1799 struct mount *mp = dvp->v_mount;
1800 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1801 int error;
1802
1803 PUFFS_MSG_ALLOC(vn, remove);
1804 remove_msg->pvnr_cookie_targ = VPTOPNC(vp);
1805 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1806 cnp, PUFFS_USE_FULLPNBUF(pmp));
1807 puffs_msg_setinfo(park_remove, PUFFSOP_VN,
1808 PUFFS_VN_REMOVE, VPTOPNC(dvp));
1809
1810 puffs_msg_enqueue(pmp, park_remove);
1811 REFPN_AND_UNLOCKVP(dvp, dpn);
1812 if (dvp == vp)
1813 REFPN(pn);
1814 else
1815 REFPN_AND_UNLOCKVP(vp, pn);
1816 error = puffs_msg_wait2(pmp, park_remove, dpn, pn);
1817
1818 PUFFS_MSG_RELEASE(remove);
1819
1820 RELEPN_AND_VP(dvp, dpn);
1821 RELEPN_AND_VP(vp, pn);
1822
1823 error = checkerr(pmp, error, __func__);
1824 return error;
1825 }
1826
1827 int
1828 puffs_vnop_mkdir(void *v)
1829 {
1830 struct vop_mkdir_v3_args /* {
1831 const struct vnodeop_desc *a_desc;
1832 struct vnode *a_dvp;
1833 struct vnode **a_vpp;
1834 struct componentname *a_cnp;
1835 struct vattr *a_vap;
1836 } */ *ap = v;
1837 PUFFS_MSG_VARS(vn, mkdir);
1838 struct vnode *dvp = ap->a_dvp;
1839 struct puffs_node *dpn = VPTOPP(dvp);
1840 struct componentname *cnp = ap->a_cnp;
1841 struct mount *mp = dvp->v_mount;
1842 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1843 int error;
1844
1845 PUFFS_MSG_ALLOC(vn, mkdir);
1846 puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred,
1847 cnp, PUFFS_USE_FULLPNBUF(pmp));
1848 mkdir_msg->pvnr_va = *ap->a_vap;
1849 puffs_msg_setinfo(park_mkdir, PUFFSOP_VN,
1850 PUFFS_VN_MKDIR, VPTOPNC(dvp));
1851
1852 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error);
1853
1854 error = checkerr(pmp, error, __func__);
1855 if (error)
1856 goto out;
1857
1858 error = puffs_newnode(mp, dvp, ap->a_vpp,
1859 mkdir_msg->pvnr_newnode, cnp, VDIR, 0);
1860 if (error) {
1861 puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie,
1862 mkdir_msg->pvnr_newnode, cnp);
1863 goto out;
1864 }
1865
1866 if (PUFFS_USE_FS_TTL(pmp)) {
1867 struct timespec *va_ttl = &mkdir_msg->pvnr_va_ttl;
1868 struct timespec *cn_ttl = &mkdir_msg->pvnr_cn_ttl;
1869 struct vattr *rvap = &mkdir_msg->pvnr_va;
1870
1871 update_va(*ap->a_vpp, NULL, rvap,
1872 va_ttl, cn_ttl, SETATTR_CHSIZE);
1873 }
1874
1875 VPTOPP(*ap->a_vpp)->pn_nlookup++;
1876
1877 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
1878 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
1879 update_parent(*ap->a_vpp, dvp);
1880
1881 out:
1882 PUFFS_MSG_RELEASE(mkdir);
1883 return error;
1884 }
1885
1886 static int
1887 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1888 struct componentname *cnp)
1889 {
1890 PUFFS_MSG_VARS(vn, rmdir);
1891 int error;
1892
1893 PUFFS_MSG_ALLOC(vn, rmdir);
1894 rmdir_msg->pvnr_cookie_targ = ck;
1895 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1896 cnp, PUFFS_USE_FULLPNBUF(pmp));
1897 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1898
1899 PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1900 PUFFS_MSG_RELEASE(rmdir);
1901
1902 return checkerr(pmp, error, __func__);
1903 }
1904
1905 int
1906 puffs_vnop_rmdir(void *v)
1907 {
1908 struct vop_rmdir_args /* {
1909 const struct vnodeop_desc *a_desc;
1910 struct vnode *a_dvp;
1911 struct vnode *a_vp;
1912 struct componentname *a_cnp;
1913 } */ *ap = v;
1914 PUFFS_MSG_VARS(vn, rmdir);
1915 struct vnode *dvp = ap->a_dvp;
1916 struct vnode *vp = ap->a_vp;
1917 struct puffs_node *dpn = VPTOPP(dvp);
1918 struct puffs_node *pn = VPTOPP(vp);
1919 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1920 struct componentname *cnp = ap->a_cnp;
1921 int error;
1922
1923 PUFFS_MSG_ALLOC(vn, rmdir);
1924 rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1925 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1926 cnp, PUFFS_USE_FULLPNBUF(pmp));
1927 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1928 PUFFS_VN_RMDIR, VPTOPNC(dvp));
1929
1930 puffs_msg_enqueue(pmp, park_rmdir);
1931 REFPN_AND_UNLOCKVP(dvp, dpn);
1932 REFPN_AND_UNLOCKVP(vp, pn);
1933 error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
1934
1935 PUFFS_MSG_RELEASE(rmdir);
1936
1937 /* XXX: some call cache_purge() *for both vnodes* here, investigate */
1938 RELEPN_AND_VP(dvp, dpn);
1939 RELEPN_AND_VP(vp, pn);
1940
1941 return error;
1942 }
1943
1944 int
1945 puffs_vnop_link(void *v)
1946 {
1947 struct vop_link_args /* {
1948 const struct vnodeop_desc *a_desc;
1949 struct vnode *a_dvp;
1950 struct vnode *a_vp;
1951 struct componentname *a_cnp;
1952 } */ *ap = v;
1953 PUFFS_MSG_VARS(vn, link);
1954 struct vnode *dvp = ap->a_dvp;
1955 struct vnode *vp = ap->a_vp;
1956 struct puffs_node *dpn = VPTOPP(dvp);
1957 struct puffs_node *pn = VPTOPP(vp);
1958 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1959 struct componentname *cnp = ap->a_cnp;
1960 int error;
1961
1962 PUFFS_MSG_ALLOC(vn, link);
1963 link_msg->pvnr_cookie_targ = VPTOPNC(vp);
1964 puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
1965 cnp, PUFFS_USE_FULLPNBUF(pmp));
1966 puffs_msg_setinfo(park_link, PUFFSOP_VN,
1967 PUFFS_VN_LINK, VPTOPNC(dvp));
1968
1969 puffs_msg_enqueue(pmp, park_link);
1970 REFPN_AND_UNLOCKVP(dvp, dpn);
1971 REFPN(pn);
1972 error = puffs_msg_wait2(pmp, park_link, dpn, pn);
1973
1974 PUFFS_MSG_RELEASE(link);
1975
1976 error = checkerr(pmp, error, __func__);
1977
1978 /*
1979 * XXX: stay in touch with the cache. I don't like this, but
1980 * don't have a better solution either. See also puffs_rename().
1981 */
1982 if (error == 0)
1983 puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
1984
1985 RELEPN_AND_VP(dvp, dpn);
1986 puffs_releasenode(pn);
1987
1988 return error;
1989 }
1990
1991 int
1992 puffs_vnop_symlink(void *v)
1993 {
1994 struct vop_symlink_v3_args /* {
1995 const struct vnodeop_desc *a_desc;
1996 struct vnode *a_dvp;
1997 struct vnode **a_vpp;
1998 struct componentname *a_cnp;
1999 struct vattr *a_vap;
2000 char *a_target;
2001 } */ *ap = v;
2002 PUFFS_MSG_VARS(vn, symlink);
2003 struct vnode *dvp = ap->a_dvp;
2004 struct puffs_node *dpn = VPTOPP(dvp);
2005 struct mount *mp = dvp->v_mount;
2006 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2007 struct componentname *cnp = ap->a_cnp;
2008 int error;
2009
2010 *ap->a_vpp = NULL;
2011
2012 PUFFS_MSG_ALLOC(vn, symlink);
2013 puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
2014 cnp, PUFFS_USE_FULLPNBUF(pmp));
2015 symlink_msg->pvnr_va = *ap->a_vap;
2016 (void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
2017 sizeof(symlink_msg->pvnr_link));
2018 puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
2019 PUFFS_VN_SYMLINK, VPTOPNC(dvp));
2020
2021 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
2022
2023 error = checkerr(pmp, error, __func__);
2024 if (error)
2025 goto out;
2026
2027 error = puffs_newnode(mp, dvp, ap->a_vpp,
2028 symlink_msg->pvnr_newnode, cnp, VLNK, 0);
2029 if (error) {
2030 puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
2031 symlink_msg->pvnr_newnode, cnp);
2032 goto out;
2033 }
2034
2035 if (PUFFS_USE_FS_TTL(pmp)) {
2036 struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl;
2037 struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl;
2038 struct vattr *rvap = &symlink_msg->pvnr_va;
2039
2040 update_va(*ap->a_vpp, NULL, rvap,
2041 va_ttl, cn_ttl, SETATTR_CHSIZE);
2042 }
2043
2044 VPTOPP(*ap->a_vpp)->pn_nlookup++;
2045
2046 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2047 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
2048 update_parent(*ap->a_vpp, dvp);
2049
2050 out:
2051 PUFFS_MSG_RELEASE(symlink);
2052
2053 return error;
2054 }
2055
2056 int
2057 puffs_vnop_readlink(void *v)
2058 {
2059 struct vop_readlink_args /* {
2060 const struct vnodeop_desc *a_desc;
2061 struct vnode *a_vp;
2062 struct uio *a_uio;
2063 kauth_cred_t a_cred;
2064 } */ *ap = v;
2065 PUFFS_MSG_VARS(vn, readlink);
2066 struct vnode *vp = ap->a_vp;
2067 struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2068 size_t linklen;
2069 int error;
2070
2071 PUFFS_MSG_ALLOC(vn, readlink);
2072 puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
2073 linklen = sizeof(readlink_msg->pvnr_link);
2074 readlink_msg->pvnr_linklen = linklen;
2075 puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
2076 PUFFS_VN_READLINK, VPTOPNC(vp));
2077
2078 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
2079 error = checkerr(pmp, error, __func__);
2080 if (error)
2081 goto out;
2082
2083 /* bad bad user file server */
2084 if (readlink_msg->pvnr_linklen > linklen) {
2085 puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
2086 "linklen too big", VPTOPNC(ap->a_vp));
2087 error = EPROTO;
2088 goto out;
2089 }
2090
2091 error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
2092 ap->a_uio);
2093 out:
2094 PUFFS_MSG_RELEASE(readlink);
2095 return error;
2096 }
2097
2098 int
2099 puffs_vnop_rename(void *v)
2100 {
2101 struct vop_rename_args /* {
2102 const struct vnodeop_desc *a_desc;
2103 struct vnode *a_fdvp;
2104 struct vnode *a_fvp;
2105 struct componentname *a_fcnp;
2106 struct vnode *a_tdvp;
2107 struct vnode *a_tvp;
2108 struct componentname *a_tcnp;
2109 } */ *ap = v;
2110 PUFFS_MSG_VARS(vn, rename);
2111 struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
2112 struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
2113 struct puffs_node *fpn = ap->a_fvp->v_data;
2114 struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
2115 int error;
2116 bool doabort = true;
2117
2118 if ((fvp->v_mount != tdvp->v_mount) ||
2119 (tvp && (fvp->v_mount != tvp->v_mount))) {
2120 ERROUT(EXDEV);
2121 }
2122
2123 PUFFS_MSG_ALLOC(vn, rename);
2124 rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
2125 rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
2126 if (tvp)
2127 rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
2128 else
2129 rename_msg->pvnr_cookie_targ = NULL;
2130 puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
2131 ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
2132 puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
2133 ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
2134 puffs_msg_setinfo(park_rename, PUFFSOP_VN,
2135 PUFFS_VN_RENAME, VPTOPNC(fdvp));
2136
2137 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
2138 doabort = false;
2139 PUFFS_MSG_RELEASE(rename);
2140 error = checkerr(pmp, error, __func__);
2141
2142 /*
2143 * XXX: stay in touch with the cache. I don't like this, but
2144 * don't have a better solution either. See also puffs_link().
2145 */
2146 if (error == 0) {
2147 puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
2148
2149 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2150 (VPTOPP(fvp)->pn_parent != tdvp))
2151 update_parent(fvp, tdvp);
2152 }
2153
2154
2155 out:
2156 if (doabort)
2157 VOP_ABORTOP(tdvp, ap->a_tcnp);
2158 if (tvp != NULL)
2159 vput(tvp);
2160 if (tdvp == tvp)
2161 vrele(tdvp);
2162 else
2163 vput(tdvp);
2164
2165 if (doabort)
2166 VOP_ABORTOP(fdvp, ap->a_fcnp);
2167 vrele(fdvp);
2168 vrele(fvp);
2169
2170 return error;
2171 }
2172
2173 #define RWARGS(cont, iofl, move, offset, creds) \
2174 (cont)->pvnr_ioflag = (iofl); \
2175 (cont)->pvnr_resid = (move); \
2176 (cont)->pvnr_offset = (offset); \
2177 puffs_credcvt(&(cont)->pvnr_cred, creds)
2178
2179 int
2180 puffs_vnop_read(void *v)
2181 {
2182 struct vop_read_args /* {
2183 const struct vnodeop_desc *a_desc;
2184 struct vnode *a_vp;
2185 struct uio *a_uio;
2186 int a_ioflag;
2187 kauth_cred_t a_cred;
2188 } */ *ap = v;
2189 PUFFS_MSG_VARS(vn, read);
2190 struct vnode *vp = ap->a_vp;
2191 struct puffs_node *pn = VPTOPP(vp);
2192 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2193 struct uio *uio = ap->a_uio;
2194 size_t tomove, argsize;
2195 vsize_t bytelen;
2196 int error;
2197
2198 read_msg = NULL;
2199 error = 0;
2200
2201 /* std sanity */
2202 if (uio->uio_resid == 0)
2203 return 0;
2204 if (uio->uio_offset < 0)
2205 return EINVAL;
2206
2207 if (vp->v_type == VREG &&
2208 PUFFS_USE_PAGECACHE(pmp) &&
2209 !(pn->pn_stat & PNODE_RDIRECT)) {
2210 const int advice = IO_ADV_DECODE(ap->a_ioflag);
2211
2212 while (uio->uio_resid > 0) {
2213 if (vp->v_size <= uio->uio_offset) {
2214 break;
2215 }
2216 bytelen = MIN(uio->uio_resid,
2217 vp->v_size - uio->uio_offset);
2218 if (bytelen == 0)
2219 break;
2220
2221 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
2222 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
2223 if (error)
2224 break;
2225 }
2226
2227 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
2228 puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
2229 } else {
2230 /*
2231 * in case it's not a regular file or we're operating
2232 * uncached, do read in the old-fashioned style,
2233 * i.e. explicit read operations
2234 */
2235
2236 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2237 argsize = sizeof(struct puffs_vnmsg_read);
2238 puffs_msgmem_alloc(argsize + tomove, &park_read,
2239 (void *)&read_msg, 1);
2240
2241 error = 0;
2242 while (uio->uio_resid > 0) {
2243 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2244 memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
2245 RWARGS(read_msg, ap->a_ioflag, tomove,
2246 uio->uio_offset, ap->a_cred);
2247 puffs_msg_setinfo(park_read, PUFFSOP_VN,
2248 PUFFS_VN_READ, VPTOPNC(vp));
2249 puffs_msg_setdelta(park_read, tomove);
2250
2251 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
2252 NULL, error);
2253 error = checkerr(pmp, error, __func__);
2254 if (error)
2255 break;
2256
2257 if (read_msg->pvnr_resid > tomove) {
2258 puffs_senderr(pmp, PUFFS_ERR_READ,
2259 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2260 error = EPROTO;
2261 break;
2262 }
2263
2264 error = uiomove(read_msg->pvnr_data,
2265 tomove - read_msg->pvnr_resid, uio);
2266
2267 /*
2268 * in case the file is out of juice, resid from
2269 * userspace is != 0. and the error-case is
2270 * quite obvious
2271 */
2272 if (error || read_msg->pvnr_resid)
2273 break;
2274 }
2275
2276 puffs_msgmem_release(park_read);
2277 }
2278
2279 return error;
2280 }
2281
2282 /*
2283 * XXX: in case of a failure, this leaves uio in a bad state.
2284 * We could theoretically copy the uio and iovecs and "replay"
2285 * them the right amount after the userspace trip, but don't
2286 * bother for now.
2287 */
2288 int
2289 puffs_vnop_write(void *v)
2290 {
2291 struct vop_write_args /* {
2292 const struct vnodeop_desc *a_desc;
2293 struct vnode *a_vp;
2294 struct uio *a_uio;
2295 int a_ioflag;
2296 kauth_cred_t a_cred;
2297 } */ *ap = v;
2298 PUFFS_MSG_VARS(vn, write);
2299 struct vnode *vp = ap->a_vp;
2300 struct puffs_node *pn = VPTOPP(vp);
2301 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2302 struct uio *uio = ap->a_uio;
2303 size_t tomove, argsize;
2304 off_t oldoff, newoff, origoff;
2305 vsize_t bytelen;
2306 int error, uflags;
2307 int ubcflags;
2308
2309 error = uflags = 0;
2310 write_msg = NULL;
2311
2312 mutex_enter(&pn->pn_sizemtx);
2313
2314 if (vp->v_type == VREG &&
2315 PUFFS_USE_PAGECACHE(pmp) &&
2316 !(pn->pn_stat & PNODE_WDIRECT)) {
2317 ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp);
2318
2319 /*
2320 * userspace *should* be allowed to control this,
2321 * but with UBC it's a bit unclear how to handle it
2322 */
2323 if (ap->a_ioflag & IO_APPEND)
2324 uio->uio_offset = vp->v_size;
2325
2326 origoff = uio->uio_offset;
2327 while (uio->uio_resid > 0) {
2328 if (vp->v_mount->mnt_flag & MNT_RELATIME)
2329 uflags |= PUFFS_UPDATEATIME;
2330 uflags |= PUFFS_UPDATECTIME;
2331 uflags |= PUFFS_UPDATEMTIME;
2332 oldoff = uio->uio_offset;
2333 bytelen = uio->uio_resid;
2334
2335 newoff = oldoff + bytelen;
2336 if (vp->v_size < newoff) {
2337 uvm_vnp_setwritesize(vp, newoff);
2338 }
2339 error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
2340 UVM_ADV_RANDOM, ubcflags);
2341
2342 /*
2343 * In case of a ubc_uiomove() error,
2344 * opt to not extend the file at all and
2345 * return an error. Otherwise, if we attempt
2346 * to clear the memory we couldn't fault to,
2347 * we might generate a kernel page fault.
2348 */
2349 if (vp->v_size < newoff) {
2350 if (error == 0) {
2351 uflags |= PUFFS_UPDATESIZE;
2352 uvm_vnp_setsize(vp, newoff);
2353 } else {
2354 uvm_vnp_setwritesize(vp, vp->v_size);
2355 }
2356 }
2357 if (error)
2358 break;
2359
2360 /*
2361 * If we're writing large files, flush to file server
2362 * every 64k. Otherwise we can very easily exhaust
2363 * kernel and user memory, as the file server cannot
2364 * really keep up with our writing speed.
2365 *
2366 * Note: this does *NOT* honor MNT_ASYNC, because
2367 * that gives userland too much say in the kernel.
2368 */
2369 if (oldoff >> 16 != uio->uio_offset >> 16) {
2370 mutex_enter(vp->v_interlock);
2371 error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
2372 uio->uio_offset & ~0xffff,
2373 PGO_CLEANIT | PGO_SYNCIO);
2374 if (error)
2375 break;
2376 }
2377 }
2378
2379 /* synchronous I/O? */
2380 if (error == 0 && ap->a_ioflag & IO_SYNC) {
2381 mutex_enter(vp->v_interlock);
2382 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2383 round_page(uio->uio_offset),
2384 PGO_CLEANIT | PGO_SYNCIO);
2385
2386 /* write through page cache? */
2387 } else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
2388 mutex_enter(vp->v_interlock);
2389 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2390 round_page(uio->uio_offset), PGO_CLEANIT);
2391 }
2392
2393 puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
2394 } else {
2395 /* tomove is non-increasing */
2396 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2397 argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2398 puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2399
2400 while (uio->uio_resid > 0) {
2401 /* move data to buffer */
2402 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2403 memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2404 RWARGS(write_msg, ap->a_ioflag, tomove,
2405 uio->uio_offset, ap->a_cred);
2406 error = uiomove(write_msg->pvnr_data, tomove, uio);
2407 if (error)
2408 break;
2409
2410 /* move buffer to userspace */
2411 puffs_msg_setinfo(park_write, PUFFSOP_VN,
2412 PUFFS_VN_WRITE, VPTOPNC(vp));
2413 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2414 NULL, error);
2415 error = checkerr(pmp, error, __func__);
2416 if (error)
2417 break;
2418
2419 if (write_msg->pvnr_resid > tomove) {
2420 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2421 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2422 error = EPROTO;
2423 break;
2424 }
2425
2426 /* adjust file size */
2427 if (vp->v_size < uio->uio_offset)
2428 uvm_vnp_setsize(vp, uio->uio_offset);
2429
2430 /* didn't move everything? bad userspace. bail */
2431 if (write_msg->pvnr_resid != 0) {
2432 error = EIO;
2433 break;
2434 }
2435 }
2436 puffs_msgmem_release(park_write);
2437 }
2438
2439 mutex_exit(&pn->pn_sizemtx);
2440 return error;
2441 }
2442
2443 int
2444 puffs_vnop_print(void *v)
2445 {
2446 struct vop_print_args /* {
2447 struct vnode *a_vp;
2448 } */ *ap = v;
2449 PUFFS_MSG_VARS(vn, print);
2450 struct vnode *vp = ap->a_vp;
2451 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2452 struct puffs_node *pn = vp->v_data;
2453
2454 /* kernel portion */
2455 printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2456 "\tuserspace cookie: %p", vp, pn, pn->pn_cookie);
2457 if (vp->v_type == VFIFO)
2458 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
2459 printf("\n");
2460
2461 /* userspace portion */
2462 if (EXISTSOP(pmp, PRINT)) {
2463 PUFFS_MSG_ALLOC(vn, print);
2464 puffs_msg_setinfo(park_print, PUFFSOP_VN,
2465 PUFFS_VN_PRINT, VPTOPNC(vp));
2466 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_print, vp->v_data,
2467 NULL);
2468 PUFFS_MSG_RELEASE(print);
2469 }
2470
2471 return 0;
2472 }
2473
2474 int
2475 puffs_vnop_pathconf(void *v)
2476 {
2477 struct vop_pathconf_args /* {
2478 const struct vnodeop_desc *a_desc;
2479 struct vnode *a_vp;
2480 int a_name;
2481 register_t *a_retval;
2482 } */ *ap = v;
2483 PUFFS_MSG_VARS(vn, pathconf);
2484 struct vnode *vp = ap->a_vp;
2485 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2486 int error;
2487
2488 PUFFS_MSG_ALLOC(vn, pathconf);
2489 pathconf_msg->pvnr_name = ap->a_name;
2490 puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2491 PUFFS_VN_PATHCONF, VPTOPNC(vp));
2492 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2493 error = checkerr(pmp, error, __func__);
2494 if (!error)
2495 *ap->a_retval = pathconf_msg->pvnr_retval;
2496 PUFFS_MSG_RELEASE(pathconf);
2497
2498 return error;
2499 }
2500
2501 int
2502 puffs_vnop_advlock(void *v)
2503 {
2504 struct vop_advlock_args /* {
2505 const struct vnodeop_desc *a_desc;
2506 struct vnode *a_vp;
2507 void *a_id;
2508 int a_op;
2509 struct flock *a_fl;
2510 int a_flags;
2511 } */ *ap = v;
2512 PUFFS_MSG_VARS(vn, advlock);
2513 struct vnode *vp = ap->a_vp;
2514 struct puffs_node *pn = VPTOPP(vp);
2515 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2516 int error;
2517
2518 if (!EXISTSOP(pmp, ADVLOCK))
2519 return lf_advlock(ap, &pn->pn_lockf, vp->v_size);
2520
2521 PUFFS_MSG_ALLOC(vn, advlock);
2522 (void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl,
2523 sizeof(advlock_msg->pvnr_fl));
2524 advlock_msg->pvnr_id = ap->a_id;
2525 advlock_msg->pvnr_op = ap->a_op;
2526 advlock_msg->pvnr_flags = ap->a_flags;
2527 puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2528 PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2529 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2530 error = checkerr(pmp, error, __func__);
2531 PUFFS_MSG_RELEASE(advlock);
2532
2533 return error;
2534 }
2535
2536 int
2537 puffs_vnop_abortop(void *v)
2538 {
2539 struct vop_abortop_args /* {
2540 struct vnode *a_dvp;
2541 struct componentname *a_cnp;
2542 }; */ *ap = v;
2543 PUFFS_MSG_VARS(vn, abortop);
2544 struct vnode *dvp = ap->a_dvp;
2545 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2546 struct componentname *cnp = ap->a_cnp;
2547
2548 if (EXISTSOP(pmp, ABORTOP)) {
2549 PUFFS_MSG_ALLOC(vn, abortop);
2550 puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2551 cnp, PUFFS_USE_FULLPNBUF(pmp));
2552 puffs_msg_setfaf(park_abortop);
2553 puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2554 PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2555
2556 puffs_msg_enqueue(pmp, park_abortop);
2557 PUFFS_MSG_RELEASE(abortop);
2558 }
2559
2560 return genfs_abortop(v);
2561 }
2562
2563 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2564
2565 /*
2566 * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2567 */
2568 int
2569 puffs_vnop_strategy(void *v)
2570 {
2571 struct vop_strategy_args /* {
2572 const struct vnodeop_desc *a_desc;
2573 struct vnode *a_vp;
2574 struct buf *a_bp;
2575 } */ *ap = v;
2576 PUFFS_MSG_VARS(vn, rw);
2577 struct vnode *vp = ap->a_vp;
2578 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2579 struct puffs_node *pn;
2580 struct buf *bp;
2581 size_t argsize;
2582 size_t tomove, moved;
2583 int error, dofaf, cansleep, dobiodone;
2584
2585 pmp = MPTOPUFFSMP(vp->v_mount);
2586 bp = ap->a_bp;
2587 error = 0;
2588 dofaf = 0;
2589 cansleep = 0;
2590 pn = VPTOPP(vp);
2591 park_rw = NULL; /* explicit */
2592 dobiodone = 1;
2593
2594 if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2595 || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2596 ERROUT(EOPNOTSUPP);
2597
2598 /*
2599 * Short-circuit optimization: don't flush buffer in between
2600 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2601 */
2602 if (pn->pn_stat & PNODE_DYING) {
2603 KASSERT(BUF_ISWRITE(bp));
2604 bp->b_resid = 0;
2605 goto out;
2606 }
2607
2608 #ifdef DIAGNOSTIC
2609 if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2610 panic("puffs_strategy: wildly inappropriate buf bcount %d",
2611 bp->b_bcount);
2612 #endif
2613
2614 /*
2615 * See explanation for the necessity of a FAF in puffs_fsync.
2616 *
2617 * Also, do FAF in case we're suspending.
2618 * See puffs_vfsops.c:pageflush()
2619 */
2620 if (BUF_ISWRITE(bp)) {
2621 mutex_enter(vp->v_interlock);
2622 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
2623 dofaf = 1;
2624 if (pn->pn_stat & PNODE_FAF)
2625 dofaf = 1;
2626 mutex_exit(vp->v_interlock);
2627 }
2628
2629 cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1;
2630
2631 KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp));
2632
2633 /* allocate transport structure */
2634 tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2635 argsize = sizeof(struct puffs_vnmsg_rw);
2636 error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2637 (void *)&rw_msg, cansleep);
2638 if (error)
2639 goto out;
2640 RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2641
2642 /* 2x2 cases: read/write, faf/nofaf */
2643 if (BUF_ISREAD(bp)) {
2644 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2645 PUFFS_VN_READ, VPTOPNC(vp));
2646 puffs_msg_setdelta(park_rw, tomove);
2647 if (BIOASYNC(bp)) {
2648 puffs_msg_setcall(park_rw,
2649 puffs_parkdone_asyncbioread, bp);
2650 puffs_msg_enqueue(pmp, park_rw);
2651 dobiodone = 0;
2652 } else {
2653 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2654 NULL, error);
2655 error = checkerr(pmp, error, __func__);
2656 if (error)
2657 goto out;
2658
2659 if (rw_msg->pvnr_resid > tomove) {
2660 puffs_senderr(pmp, PUFFS_ERR_READ,
2661 E2BIG, "resid grew", VPTOPNC(vp));
2662 ERROUT(EPROTO);
2663 }
2664
2665 moved = tomove - rw_msg->pvnr_resid;
2666
2667 (void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2668 bp->b_resid = bp->b_bcount - moved;
2669 }
2670 } else {
2671 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2672 PUFFS_VN_WRITE, VPTOPNC(vp));
2673 /*
2674 * make pages read-only before we write them if we want
2675 * write caching info
2676 */
2677 if (PUFFS_WCACHEINFO(pmp)) {
2678 struct uvm_object *uobj = &vp->v_uobj;
2679 int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2680 struct vm_page *vmp;
2681 int i;
2682
2683 for (i = 0; i < npages; i++) {
2684 vmp= uvm_pageratop((vaddr_t)bp->b_data
2685 + (i << PAGE_SHIFT));
2686 DPRINTF(("puffs_strategy: write-protecting "
2687 "vp %p page %p, offset %" PRId64"\n",
2688 vp, vmp, vmp->offset));
2689 mutex_enter(uobj->vmobjlock);
2690 vmp->flags |= PG_RDONLY;
2691 pmap_page_protect(vmp, VM_PROT_READ);
2692 mutex_exit(uobj->vmobjlock);
2693 }
2694 }
2695
2696 (void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2697 if (dofaf) {
2698 puffs_msg_setfaf(park_rw);
2699 } else if (BIOASYNC(bp)) {
2700 puffs_msg_setcall(park_rw,
2701 puffs_parkdone_asyncbiowrite, bp);
2702 dobiodone = 0;
2703 }
2704
2705 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2706
2707 if (dobiodone == 0)
2708 goto out;
2709
2710 /*
2711 * XXXXXXXX: wrong, but kernel can't survive strategy
2712 * failure currently. Here, have one more X: X.
2713 */
2714 if (error != ENOMEM)
2715 error = 0;
2716
2717 error = checkerr(pmp, error, __func__);
2718 if (error)
2719 goto out;
2720
2721 if (rw_msg->pvnr_resid > tomove) {
2722 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2723 E2BIG, "resid grew", VPTOPNC(vp));
2724 ERROUT(EPROTO);
2725 }
2726
2727 /*
2728 * FAF moved everything. Frankly, we don't
2729 * really have a choice.
2730 */
2731 if (dofaf && error == 0)
2732 moved = tomove;
2733 else
2734 moved = tomove - rw_msg->pvnr_resid;
2735
2736 bp->b_resid = bp->b_bcount - moved;
2737 if (bp->b_resid != 0) {
2738 ERROUT(EIO);
2739 }
2740 }
2741
2742 out:
2743 if (park_rw)
2744 puffs_msgmem_release(park_rw);
2745
2746 if (error)
2747 bp->b_error = error;
2748
2749 if (error || dobiodone)
2750 biodone(bp);
2751
2752 return error;
2753 }
2754
2755 int
2756 puffs_vnop_mmap(void *v)
2757 {
2758 struct vop_mmap_args /* {
2759 const struct vnodeop_desc *a_desc;
2760 struct vnode *a_vp;
2761 vm_prot_t a_prot;
2762 kauth_cred_t a_cred;
2763 } */ *ap = v;
2764 PUFFS_MSG_VARS(vn, mmap);
2765 struct vnode *vp = ap->a_vp;
2766 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2767 int error;
2768
2769 if (!PUFFS_USE_PAGECACHE(pmp))
2770 return genfs_eopnotsupp(v);
2771
2772 if (EXISTSOP(pmp, MMAP)) {
2773 PUFFS_MSG_ALLOC(vn, mmap);
2774 mmap_msg->pvnr_prot = ap->a_prot;
2775 puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2776 puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2777 PUFFS_VN_MMAP, VPTOPNC(vp));
2778
2779 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2780 error = checkerr(pmp, error, __func__);
2781 PUFFS_MSG_RELEASE(mmap);
2782 } else {
2783 error = genfs_mmap(v);
2784 }
2785
2786 return error;
2787 }
2788
2789
2790 /*
2791 * The rest don't get a free trip to userspace and back, they
2792 * have to stay within the kernel.
2793 */
2794
2795 /*
2796 * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2797 * well, maybe somehow, somewhere, some day ....
2798 */
2799 int
2800 puffs_vnop_bmap(void *v)
2801 {
2802 struct vop_bmap_args /* {
2803 const struct vnodeop_desc *a_desc;
2804 struct vnode *a_vp;
2805 daddr_t a_bn;
2806 struct vnode **a_vpp;
2807 daddr_t *a_bnp;
2808 int *a_runp;
2809 } */ *ap = v;
2810 struct puffs_mount *pmp;
2811
2812 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2813
2814 if (ap->a_vpp)
2815 *ap->a_vpp = ap->a_vp;
2816 if (ap->a_bnp)
2817 *ap->a_bnp = ap->a_bn;
2818 if (ap->a_runp)
2819 *ap->a_runp
2820 = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
2821
2822 return 0;
2823 }
2824
2825 /*
2826 * Handle getpages faults in puffs. We let genfs_getpages() do most
2827 * of the dirty work, but we come in this route to do accounting tasks.
2828 * If the user server has specified functions for cache notifications
2829 * about reads and/or writes, we record which type of operation we got,
2830 * for which page range, and proceed to issue a FAF notification to the
2831 * server about it.
2832 */
2833 int
2834 puffs_vnop_getpages(void *v)
2835 {
2836 struct vop_getpages_args /* {
2837 const struct vnodeop_desc *a_desc;
2838 struct vnode *a_vp;
2839 voff_t a_offset;
2840 struct vm_page **a_m;
2841 int *a_count;
2842 int a_centeridx;
2843 vm_prot_t a_access_type;
2844 int a_advice;
2845 int a_flags;
2846 } */ *ap = v;
2847 struct puffs_mount *pmp;
2848 struct puffs_node *pn;
2849 struct vnode *vp;
2850 struct vm_page **pgs;
2851 struct puffs_cacheinfo *pcinfo = NULL;
2852 struct puffs_cacherun *pcrun;
2853 void *parkmem = NULL;
2854 size_t runsizes;
2855 int i, npages, si, streakon;
2856 int error, locked, write;
2857
2858 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2859 npages = *ap->a_count;
2860 pgs = ap->a_m;
2861 vp = ap->a_vp;
2862 pn = vp->v_data;
2863 locked = (ap->a_flags & PGO_LOCKED) != 0;
2864 write = (ap->a_access_type & VM_PROT_WRITE) != 0;
2865
2866 /* ccg xnaht - gets Wuninitialized wrong */
2867 pcrun = NULL;
2868 runsizes = 0;
2869
2870 /*
2871 * Check that we aren't trying to fault in pages which our file
2872 * server doesn't know about. This happens if we extend a file by
2873 * skipping some pages and later try to fault in pages which
2874 * are between pn_serversize and vp_size. This check optimizes
2875 * away the common case where a file is being extended.
2876 */
2877 if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
2878 struct vattr va;
2879
2880 /* try again later when we can block */
2881 if (locked)
2882 ERROUT(EBUSY);
2883
2884 mutex_exit(vp->v_interlock);
2885 vattr_null(&va);
2886 va.va_size = vp->v_size;
2887 error = dosetattr(vp, &va, FSCRED, 0);
2888 if (error)
2889 ERROUT(error);
2890 mutex_enter(vp->v_interlock);
2891 }
2892
2893 if (write && PUFFS_WCACHEINFO(pmp)) {
2894 #ifdef notnowjohn
2895 /* allocate worst-case memory */
2896 runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
2897 KASSERT(curlwp != uvm.pagedaemon_lwp || locked);
2898 pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
2899 locked ? KM_NOSLEEP : KM_SLEEP);
2900
2901 /*
2902 * can't block if we're locked and can't mess up caching
2903 * information for fs server. so come back later, please
2904 */
2905 if (pcinfo == NULL)
2906 ERROUT(ENOMEM);
2907
2908 parkmem = puffs_park_alloc(locked == 0);
2909 if (parkmem == NULL)
2910 ERROUT(ENOMEM);
2911
2912 pcrun = pcinfo->pcache_runs;
2913 #else
2914 (void)parkmem;
2915 #endif
2916 }
2917
2918 error = genfs_getpages(v);
2919 if (error)
2920 goto out;
2921
2922 if (PUFFS_WCACHEINFO(pmp) == 0)
2923 goto out;
2924
2925 /*
2926 * Let's see whose fault it was and inform the user server of
2927 * possibly read/written pages. Map pages from read faults
2928 * strictly read-only, since otherwise we might miss info on
2929 * when the page is actually write-faulted to.
2930 */
2931 if (!locked)
2932 mutex_enter(vp->v_uobj.vmobjlock);
2933 for (i = 0, si = 0, streakon = 0; i < npages; i++) {
2934 if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
2935 if (streakon && write) {
2936 streakon = 0;
2937 pcrun[si].pcache_runend
2938 = trunc_page(pgs[i]->offset) + PAGE_MASK;
2939 si++;
2940 }
2941 continue;
2942 }
2943 if (streakon == 0 && write) {
2944 streakon = 1;
2945 pcrun[si].pcache_runstart = pgs[i]->offset;
2946 }
2947
2948 if (!write)
2949 pgs[i]->flags |= PG_RDONLY;
2950 }
2951 /* was the last page part of our streak? */
2952 if (streakon) {
2953 pcrun[si].pcache_runend
2954 = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
2955 si++;
2956 }
2957 if (!locked)
2958 mutex_exit(vp->v_uobj.vmobjlock);
2959
2960 KASSERT(si <= (npages / 2) + 1);
2961
2962 #ifdef notnowjohn
2963 /* send results to userspace */
2964 if (write)
2965 puffs_cacheop(pmp, parkmem, pcinfo,
2966 sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
2967 #endif
2968
2969 out:
2970 if (error) {
2971 if (pcinfo != NULL)
2972 kmem_free(pcinfo,
2973 sizeof(struct puffs_cacheinfo) + runsizes);
2974 #ifdef notnowjohn
2975 if (parkmem != NULL)
2976 puffs_park_release(parkmem, 1);
2977 #endif
2978 }
2979
2980 return error;
2981 }
2982
2983 /*
2984 * Extended attribute support.
2985 */
2986
2987 int
2988 puffs_vnop_getextattr(void *v)
2989 {
2990 struct vop_getextattr_args /*
2991 struct vnode *a_vp;
2992 int a_attrnamespace;
2993 const char *a_name;
2994 struct uio *a_uio;
2995 size_t *a_size;
2996 kauth_cred_t a_cred;
2997 }; */ *ap = v;
2998 PUFFS_MSG_VARS(vn, getextattr);
2999 struct vnode *vp = ap->a_vp;
3000 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3001 int attrnamespace = ap->a_attrnamespace;
3002 const char *name = ap->a_name;
3003 struct uio *uio = ap->a_uio;
3004 size_t *sizep = ap->a_size;
3005 size_t tomove, resid;
3006 int error;
3007
3008 if (uio)
3009 resid = uio->uio_resid;
3010 else
3011 resid = 0;
3012
3013 tomove = PUFFS_TOMOVE(resid, pmp);
3014 if (tomove != resid) {
3015 error = E2BIG;
3016 goto out;
3017 }
3018
3019 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove,
3020 &park_getextattr, (void *)&getextattr_msg, 1);
3021
3022 getextattr_msg->pvnr_attrnamespace = attrnamespace;
3023 strlcpy(getextattr_msg->pvnr_attrname, name,
3024 sizeof(getextattr_msg->pvnr_attrname));
3025 puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred);
3026 if (sizep)
3027 getextattr_msg->pvnr_datasize = 1;
3028 getextattr_msg->pvnr_resid = tomove;
3029
3030 puffs_msg_setinfo(park_getextattr,
3031 PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp));
3032 puffs_msg_setdelta(park_getextattr, tomove);
3033 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error);
3034
3035 error = checkerr(pmp, error, __func__);
3036 if (error)
3037 goto out;
3038
3039 resid = getextattr_msg->pvnr_resid;
3040 if (resid > tomove) {
3041 puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG,
3042 "resid grew", VPTOPNC(vp));
3043 error = EPROTO;
3044 goto out;
3045 }
3046
3047 if (sizep)
3048 *sizep = getextattr_msg->pvnr_datasize;
3049 if (uio)
3050 error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio);
3051
3052 out:
3053 PUFFS_MSG_RELEASE(getextattr);
3054 return error;
3055 }
3056
3057 int
3058 puffs_vnop_setextattr(void *v)
3059 {
3060 struct vop_setextattr_args /* {
3061 struct vnode *a_vp;
3062 int a_attrnamespace;
3063 const char *a_name;
3064 struct uio *a_uio;
3065 kauth_cred_t a_cred;
3066 }; */ *ap = v;
3067 PUFFS_MSG_VARS(vn, setextattr);
3068 struct vnode *vp = ap->a_vp;
3069 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3070 int attrnamespace = ap->a_attrnamespace;
3071 const char *name = ap->a_name;
3072 struct uio *uio = ap->a_uio;
3073 size_t tomove, resid;
3074 int error;
3075
3076 if (uio)
3077 resid = uio->uio_resid;
3078 else
3079 resid = 0;
3080
3081 tomove = PUFFS_TOMOVE(resid, pmp);
3082 if (tomove != resid) {
3083 error = E2BIG;
3084 goto out;
3085 }
3086
3087 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove,
3088 &park_setextattr, (void *)&setextattr_msg, 1);
3089
3090 setextattr_msg->pvnr_attrnamespace = attrnamespace;
3091 strlcpy(setextattr_msg->pvnr_attrname, name,
3092 sizeof(setextattr_msg->pvnr_attrname));
3093 puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred);
3094 setextattr_msg->pvnr_resid = tomove;
3095
3096 if (uio) {
3097 error = uiomove(setextattr_msg->pvnr_data, tomove, uio);
3098 if (error)
3099 goto out;
3100 }
3101
3102 puffs_msg_setinfo(park_setextattr,
3103 PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp));
3104 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error);
3105
3106 error = checkerr(pmp, error, __func__);
3107 if (error)
3108 goto out;
3109
3110 if (setextattr_msg->pvnr_resid != 0)
3111 error = EIO;
3112
3113 out:
3114 PUFFS_MSG_RELEASE(setextattr);
3115
3116 return error;
3117 }
3118
3119 int
3120 puffs_vnop_listextattr(void *v)
3121 {
3122 struct vop_listextattr_args /* {
3123 struct vnode *a_vp;
3124 int a_attrnamespace;
3125 struct uio *a_uio;
3126 size_t *a_size;
3127 int a_flag,
3128 kauth_cred_t a_cred;
3129 }; */ *ap = v;
3130 PUFFS_MSG_VARS(vn, listextattr);
3131 struct vnode *vp = ap->a_vp;
3132 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3133 int attrnamespace = ap->a_attrnamespace;
3134 struct uio *uio = ap->a_uio;
3135 size_t *sizep = ap->a_size;
3136 int flag = ap->a_flag;
3137 size_t tomove, resid;
3138 int error;
3139
3140 if (uio)
3141 resid = uio->uio_resid;
3142 else
3143 resid = 0;
3144
3145 tomove = PUFFS_TOMOVE(resid, pmp);
3146 if (tomove != resid) {
3147 error = E2BIG;
3148 goto out;
3149 }
3150
3151 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_listextattr) + tomove,
3152 &park_listextattr, (void *)&listextattr_msg, 1);
3153
3154 listextattr_msg->pvnr_attrnamespace = attrnamespace;
3155 listextattr_msg->pvnr_flag = flag;
3156 puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred);
3157 listextattr_msg->pvnr_resid = tomove;
3158 if (sizep)
3159 listextattr_msg->pvnr_datasize = 1;
3160
3161 puffs_msg_setinfo(park_listextattr,
3162 PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp));
3163 puffs_msg_setdelta(park_listextattr, tomove);
3164 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error);
3165
3166 error = checkerr(pmp, error, __func__);
3167 if (error)
3168 goto out;
3169
3170 resid = listextattr_msg->pvnr_resid;
3171 if (resid > tomove) {
3172 puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG,
3173 "resid grew", VPTOPNC(vp));
3174 error = EPROTO;
3175 goto out;
3176 }
3177
3178 if (sizep)
3179 *sizep = listextattr_msg->pvnr_datasize;
3180 if (uio)
3181 error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio);
3182
3183 out:
3184 PUFFS_MSG_RELEASE(listextattr);
3185 return error;
3186 }
3187
3188 int
3189 puffs_vnop_deleteextattr(void *v)
3190 {
3191 struct vop_deleteextattr_args /* {
3192 struct vnode *a_vp;
3193 int a_attrnamespace;
3194 const char *a_name;
3195 kauth_cred_t a_cred;
3196 }; */ *ap = v;
3197 PUFFS_MSG_VARS(vn, deleteextattr);
3198 struct vnode *vp = ap->a_vp;
3199 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3200 int attrnamespace = ap->a_attrnamespace;
3201 const char *name = ap->a_name;
3202 int error;
3203
3204 PUFFS_MSG_ALLOC(vn, deleteextattr);
3205 deleteextattr_msg->pvnr_attrnamespace = attrnamespace;
3206 strlcpy(deleteextattr_msg->pvnr_attrname, name,
3207 sizeof(deleteextattr_msg->pvnr_attrname));
3208 puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred);
3209
3210 puffs_msg_setinfo(park_deleteextattr,
3211 PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp));
3212 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr,
3213 vp->v_data, NULL, error);
3214
3215 error = checkerr(pmp, error, __func__);
3216
3217 PUFFS_MSG_RELEASE(deleteextattr);
3218 return error;
3219 }
3220
3221 /*
3222 * spec & fifo. These call the miscfs spec and fifo vectors, but issue
3223 * FAF update information for the puffs node first.
3224 */
3225 int
3226 puffs_vnop_spec_read(void *v)
3227 {
3228 struct vop_read_args /* {
3229 const struct vnodeop_desc *a_desc;
3230 struct vnode *a_vp;
3231 struct uio *a_uio;
3232 int a_ioflag;
3233 kauth_cred_t a_cred;
3234 } */ *ap = v;
3235
3236 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3237 return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
3238 }
3239
3240 int
3241 puffs_vnop_spec_write(void *v)
3242 {
3243 struct vop_write_args /* {
3244 const struct vnodeop_desc *a_desc;
3245 struct vnode *a_vp;
3246 struct uio *a_uio;
3247 int a_ioflag;
3248 kauth_cred_t a_cred;
3249 } */ *ap = v;
3250
3251 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3252 return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
3253 }
3254
3255 int
3256 puffs_vnop_fifo_read(void *v)
3257 {
3258 struct vop_read_args /* {
3259 const struct vnodeop_desc *a_desc;
3260 struct vnode *a_vp;
3261 struct uio *a_uio;
3262 int a_ioflag;
3263 kauth_cred_t a_cred;
3264 } */ *ap = v;
3265
3266 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3267 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
3268 }
3269
3270 int
3271 puffs_vnop_fifo_write(void *v)
3272 {
3273 struct vop_write_args /* {
3274 const struct vnodeop_desc *a_desc;
3275 struct vnode *a_vp;
3276 struct uio *a_uio;
3277 int a_ioflag;
3278 kauth_cred_t a_cred;
3279 } */ *ap = v;
3280
3281 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3282 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
3283 }
3284