puffs_vnops.c revision 1.182 1 /* $NetBSD: puffs_vnops.c,v 1.182 2014/07/25 08:20:52 dholland 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 1.182 2014/07/25 08:20:52 dholland 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 int mode = ap->a_mode;
897 int error;
898
899 DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode));
900
901 if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE))
902 ERROUT(EROFS);
903
904 if (!EXISTSOP(pmp, OPEN))
905 ERROUT(0);
906
907 PUFFS_MSG_ALLOC(vn, open);
908 open_msg->pvnr_mode = mode;
909 puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred);
910 puffs_msg_setinfo(park_open, PUFFSOP_VN,
911 PUFFS_VN_OPEN, VPTOPNC(vp));
912
913 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error);
914 error = checkerr(pmp, error, __func__);
915
916 out:
917 DPRINTF(("puffs_open: returning %d\n", error));
918 PUFFS_MSG_RELEASE(open);
919 return error;
920 }
921
922 int
923 puffs_vnop_close(void *v)
924 {
925 struct vop_close_args /* {
926 const struct vnodeop_desc *a_desc;
927 struct vnode *a_vp;
928 int a_fflag;
929 kauth_cred_t a_cred;
930 } */ *ap = v;
931 PUFFS_MSG_VARS(vn, close);
932 struct vnode *vp = ap->a_vp;
933 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
934
935 PUFFS_MSG_ALLOC(vn, close);
936 puffs_msg_setfaf(park_close);
937 close_msg->pvnr_fflag = ap->a_fflag;
938 puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
939 puffs_msg_setinfo(park_close, PUFFSOP_VN,
940 PUFFS_VN_CLOSE, VPTOPNC(vp));
941
942 puffs_msg_enqueue(pmp, park_close);
943 PUFFS_MSG_RELEASE(close);
944 return 0;
945 }
946
947 int
948 puffs_vnop_access(void *v)
949 {
950 struct vop_access_args /* {
951 const struct vnodeop_desc *a_desc;
952 struct vnode *a_vp;
953 int a_mode;
954 kauth_cred_t a_cred;
955 } */ *ap = v;
956 PUFFS_MSG_VARS(vn, access);
957 struct vnode *vp = ap->a_vp;
958 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
959 int mode = ap->a_mode;
960 int error;
961
962 if (mode & VWRITE) {
963 switch (vp->v_type) {
964 case VDIR:
965 case VLNK:
966 case VREG:
967 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
968 || !EXISTSOP(pmp, WRITE))
969 return EROFS;
970 break;
971 default:
972 break;
973 }
974 }
975
976 if (!EXISTSOP(pmp, ACCESS))
977 return 0;
978
979 PUFFS_MSG_ALLOC(vn, access);
980 access_msg->pvnr_mode = ap->a_mode;
981 puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
982 puffs_msg_setinfo(park_access, PUFFSOP_VN,
983 PUFFS_VN_ACCESS, VPTOPNC(vp));
984
985 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
986 error = checkerr(pmp, error, __func__);
987 PUFFS_MSG_RELEASE(access);
988
989 return error;
990 }
991
992 static void
993 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap,
994 struct timespec *va_ttl, struct timespec *cn_ttl, int flags)
995 {
996 struct puffs_node *pn = VPTOPP(vp);
997
998 if (TTL_VALID(cn_ttl)) {
999 pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl);
1000 pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace);
1001 }
1002
1003 /*
1004 * Don't listen to the file server regarding special device
1005 * size info, the file server doesn't know anything about them.
1006 */
1007 if (vp->v_type == VBLK || vp->v_type == VCHR)
1008 rvap->va_size = vp->v_size;
1009
1010 /* Ditto for blocksize (ufs comment: this doesn't belong here) */
1011 if (vp->v_type == VBLK)
1012 rvap->va_blocksize = BLKDEV_IOSIZE;
1013 else if (vp->v_type == VCHR)
1014 rvap->va_blocksize = MAXBSIZE;
1015
1016 if (vap != NULL) {
1017 (void) memcpy(vap, rvap, sizeof(struct vattr));
1018 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1019
1020 if (pn->pn_stat & PNODE_METACACHE_ATIME)
1021 vap->va_atime = pn->pn_mc_atime;
1022 if (pn->pn_stat & PNODE_METACACHE_CTIME)
1023 vap->va_ctime = pn->pn_mc_ctime;
1024 if (pn->pn_stat & PNODE_METACACHE_MTIME)
1025 vap->va_mtime = pn->pn_mc_mtime;
1026 if (pn->pn_stat & PNODE_METACACHE_SIZE)
1027 vap->va_size = pn->pn_mc_size;
1028 }
1029
1030 if (!(pn->pn_stat & PNODE_METACACHE_SIZE) && (flags & SETATTR_CHSIZE)) {
1031 if (rvap->va_size != VNOVAL
1032 && vp->v_type != VBLK && vp->v_type != VCHR) {
1033 uvm_vnp_setsize(vp, rvap->va_size);
1034 pn->pn_serversize = rvap->va_size;
1035 }
1036 }
1037
1038 if ((va_ttl != NULL) && TTL_VALID(va_ttl)) {
1039 if (pn->pn_va_cache == NULL)
1040 pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK);
1041
1042 (void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap));
1043
1044 pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl);
1045 }
1046 }
1047
1048 static void
1049 update_parent(struct vnode *vp, struct vnode *dvp)
1050 {
1051 struct puffs_node *pn = VPTOPP(vp);
1052
1053 if (pn->pn_parent != NULL) {
1054 KASSERT(pn->pn_parent != dvp);
1055 vrele(pn->pn_parent);
1056 }
1057
1058 vref(dvp);
1059 pn->pn_parent = dvp;
1060 }
1061
1062 int
1063 puffs_vnop_getattr(void *v)
1064 {
1065 struct vop_getattr_args /* {
1066 const struct vnodeop_desc *a_desc;
1067 struct vnode *a_vp;
1068 struct vattr *a_vap;
1069 kauth_cred_t a_cred;
1070 } */ *ap = v;
1071 PUFFS_MSG_VARS(vn, getattr);
1072 struct vnode *vp = ap->a_vp;
1073 struct mount *mp = vp->v_mount;
1074 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1075 struct vattr *vap, *rvap;
1076 struct puffs_node *pn = VPTOPP(vp);
1077 struct timespec *va_ttl = NULL;
1078 int error = 0;
1079
1080 /*
1081 * A lock is required so that we do not race with
1082 * setattr, write and fsync when changing vp->v_size.
1083 * This is critical, since setting a stall smaler value
1084 * triggers a file truncate in uvm_vnp_setsize(), which
1085 * most of the time means data corruption (a chunk of
1086 * data is replaced by zeroes). This can be removed if
1087 * we decide one day that VOP_GETATTR must operate on
1088 * a locked vnode.
1089 *
1090 * XXX Should be useless now that VOP_GETATTR has been
1091 * fixed to always require a shared lock at least.
1092 */
1093 mutex_enter(&pn->pn_sizemtx);
1094
1095 REFPN(pn);
1096 vap = ap->a_vap;
1097
1098 if (PUFFS_USE_FS_TTL(pmp)) {
1099 if (!TIMED_OUT(pn->pn_va_timeout)) {
1100 update_va(vp, vap, pn->pn_va_cache,
1101 NULL, NULL, SETATTR_CHSIZE);
1102 goto out2;
1103 }
1104 }
1105
1106 PUFFS_MSG_ALLOC(vn, getattr);
1107 vattr_null(&getattr_msg->pvnr_va);
1108 puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
1109 puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
1110 PUFFS_VN_GETATTR, VPTOPNC(vp));
1111
1112 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
1113 error = checkerr(pmp, error, __func__);
1114 if (error)
1115 goto out;
1116
1117 rvap = &getattr_msg->pvnr_va;
1118
1119 if (PUFFS_USE_FS_TTL(pmp))
1120 va_ttl = &getattr_msg->pvnr_va_ttl;
1121
1122 update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE);
1123
1124 out:
1125 PUFFS_MSG_RELEASE(getattr);
1126
1127 out2:
1128 puffs_releasenode(pn);
1129
1130 mutex_exit(&pn->pn_sizemtx);
1131
1132 return error;
1133 }
1134
1135 static int
1136 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
1137 {
1138 PUFFS_MSG_VARS(vn, setattr);
1139 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1140 struct puffs_node *pn = vp->v_data;
1141 int error = 0;
1142
1143 KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx));
1144
1145 if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
1146 (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
1147 || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
1148 || vap->va_mode != (mode_t)VNOVAL))
1149 return EROFS;
1150
1151 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
1152 && vp->v_type == VREG && vap->va_size != VNOVAL)
1153 return EROFS;
1154
1155 /*
1156 * Flush metacache first. If we are called with some explicit
1157 * parameters, treat them as information overriding metacache
1158 * information.
1159 */
1160 if (pn->pn_stat & PNODE_METACACHE_MASK) {
1161 if ((pn->pn_stat & PNODE_METACACHE_ATIME)
1162 && vap->va_atime.tv_sec == VNOVAL)
1163 vap->va_atime = pn->pn_mc_atime;
1164 if ((pn->pn_stat & PNODE_METACACHE_CTIME)
1165 && vap->va_ctime.tv_sec == VNOVAL)
1166 vap->va_ctime = pn->pn_mc_ctime;
1167 if ((pn->pn_stat & PNODE_METACACHE_MTIME)
1168 && vap->va_mtime.tv_sec == VNOVAL)
1169 vap->va_mtime = pn->pn_mc_mtime;
1170 if ((pn->pn_stat & PNODE_METACACHE_SIZE)
1171 && vap->va_size == VNOVAL)
1172 vap->va_size = pn->pn_mc_size;
1173
1174 pn->pn_stat &= ~PNODE_METACACHE_MASK;
1175 }
1176
1177 /*
1178 * Flush attribute cache so that another thread do
1179 * not get a stale value during the operation.
1180 */
1181 if (PUFFS_USE_FS_TTL(pmp))
1182 pn->pn_va_timeout = 0;
1183
1184 PUFFS_MSG_ALLOC(vn, setattr);
1185 (void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
1186 puffs_credcvt(&setattr_msg->pvnr_cred, cred);
1187 puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
1188 PUFFS_VN_SETATTR, VPTOPNC(vp));
1189 if (flags & SETATTR_ASYNC)
1190 puffs_msg_setfaf(park_setattr);
1191
1192 puffs_msg_enqueue(pmp, park_setattr);
1193 if ((flags & SETATTR_ASYNC) == 0)
1194 error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
1195
1196 if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) {
1197 struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl;
1198 struct vattr *rvap = &setattr_msg->pvnr_va;
1199
1200 update_va(vp, NULL, rvap, va_ttl, NULL, flags);
1201 }
1202
1203 PUFFS_MSG_RELEASE(setattr);
1204 if ((flags & SETATTR_ASYNC) == 0) {
1205 error = checkerr(pmp, error, __func__);
1206 if (error)
1207 return error;
1208 } else {
1209 error = 0;
1210 }
1211
1212 if (vap->va_size != VNOVAL) {
1213 pn->pn_serversize = vap->va_size;
1214 if (flags & SETATTR_CHSIZE)
1215 uvm_vnp_setsize(vp, vap->va_size);
1216 }
1217
1218 return 0;
1219 }
1220
1221 int
1222 puffs_vnop_setattr(void *v)
1223 {
1224 struct vop_getattr_args /* {
1225 const struct vnodeop_desc *a_desc;
1226 struct vnode *a_vp;
1227 struct vattr *a_vap;
1228 kauth_cred_t a_cred;
1229 } */ *ap = v;
1230 struct puffs_node *pn = ap->a_vp->v_data;
1231 int error;
1232
1233 mutex_enter(&pn->pn_sizemtx);
1234 error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
1235 mutex_exit(&pn->pn_sizemtx);
1236
1237 return error;
1238 }
1239
1240 static __inline int
1241 doinact(struct puffs_mount *pmp, int iaflag)
1242 {
1243
1244 if (EXISTSOP(pmp, INACTIVE))
1245 if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
1246 if (iaflag || ALLOPS(pmp))
1247 return 1;
1248 else
1249 return 0;
1250 else
1251 return 1;
1252 else
1253 return 0;
1254 }
1255
1256 static void
1257 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
1258 {
1259 PUFFS_MSG_VARS(vn, inactive);
1260
1261 if (doinact(pmp, iaflag)) {
1262 PUFFS_MSG_ALLOC(vn, inactive);
1263 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1264 PUFFS_VN_INACTIVE, ck);
1265 PUFFS_MSG_ENQUEUEWAIT_NOERROR(pmp, park_inactive);
1266 PUFFS_MSG_RELEASE(inactive);
1267 }
1268 }
1269
1270 /* XXX: callinactive can't setback */
1271 int
1272 puffs_vnop_inactive(void *v)
1273 {
1274 struct vop_inactive_args /* {
1275 const struct vnodeop_desc *a_desc;
1276 struct vnode *a_vp;
1277 } */ *ap = v;
1278 PUFFS_MSG_VARS(vn, inactive);
1279 struct vnode *vp = ap->a_vp;
1280 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1281 struct puffs_node *pnode;
1282 bool recycle = false;
1283
1284 pnode = vp->v_data;
1285 mutex_enter(&pnode->pn_sizemtx);
1286
1287 if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1288 flushvncache(vp, 0, 0, false);
1289 PUFFS_MSG_ALLOC(vn, inactive);
1290 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1291 PUFFS_VN_INACTIVE, VPTOPNC(vp));
1292 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_inactive, vp->v_data,
1293 NULL);
1294 PUFFS_MSG_RELEASE(inactive);
1295 }
1296 pnode->pn_stat &= ~PNODE_DOINACT;
1297
1298 /*
1299 * file server thinks it's gone? then don't be afraid care,
1300 * node's life was already all it would ever be
1301 */
1302 if (pnode->pn_stat & PNODE_NOREFS) {
1303 pnode->pn_stat |= PNODE_DYING;
1304 recycle = true;
1305 }
1306
1307 /*
1308 * Handle node TTL.
1309 * If grace has already timed out, make it reclaimed.
1310 * Otherwise, we queue its expiration by sop thread, so
1311 * that it does not remain for ages in the freelist,
1312 * holding memory in userspace, while we will have
1313 * to look it up again anyway.
1314 */
1315 if (PUFFS_USE_FS_TTL(pmp) && !(vp->v_vflag & VV_ROOT) && !recycle) {
1316 bool incache = !TIMED_OUT(pnode->pn_cn_timeout);
1317 bool ingrace = !TIMED_OUT(pnode->pn_cn_grace);
1318 bool reclaimqueued = pnode->pn_stat & PNODE_SOPEXP;
1319
1320 if (!incache && !ingrace && !reclaimqueued) {
1321 pnode->pn_stat |= PNODE_DYING;
1322 recycle = true;
1323 }
1324
1325 if (!recycle && !reclaimqueued) {
1326 struct puffs_sopreq *psopr;
1327 int at = MAX(pnode->pn_cn_grace, pnode->pn_cn_timeout);
1328
1329 KASSERT(curlwp != uvm.pagedaemon_lwp);
1330 psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP);
1331 psopr->psopr_ck = VPTOPNC(pnode->pn_vp);
1332 psopr->psopr_sopreq = PUFFS_SOPREQ_EXPIRE;
1333 psopr->psopr_at = at;
1334
1335 mutex_enter(&pmp->pmp_sopmtx);
1336
1337 /*
1338 * If thread has disapeared, just give up. The
1339 * fs is being unmounted and the node will be
1340 * be reclaimed anyway.
1341 *
1342 * Otherwise, we queue the request but do not
1343 * immediatly signal the thread, as the node
1344 * has not been expired yet.
1345 */
1346 if (pmp->pmp_sopthrcount == 0) {
1347 kmem_free(psopr, sizeof(*psopr));
1348 } else {
1349 TAILQ_INSERT_TAIL(&pmp->pmp_sopnodereqs,
1350 psopr, psopr_entries);
1351 pnode->pn_stat |= PNODE_SOPEXP;
1352 }
1353
1354 mutex_exit(&pmp->pmp_sopmtx);
1355 }
1356 }
1357
1358 *ap->a_recycle = recycle;
1359
1360 mutex_exit(&pnode->pn_sizemtx);
1361 VOP_UNLOCK(vp);
1362
1363 return 0;
1364 }
1365
1366 static void
1367 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck, int nlookup)
1368 {
1369 PUFFS_MSG_VARS(vn, reclaim);
1370
1371 if (!EXISTSOP(pmp, RECLAIM))
1372 return;
1373
1374 PUFFS_MSG_ALLOC(vn, reclaim);
1375 reclaim_msg->pvnr_nlookup = nlookup;
1376 puffs_msg_setfaf(park_reclaim);
1377 puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck);
1378
1379 puffs_msg_enqueue(pmp, park_reclaim);
1380 PUFFS_MSG_RELEASE(reclaim);
1381 return;
1382 }
1383
1384 /*
1385 * always FAF, we don't really care if the server wants to fail to
1386 * reclaim the node or not
1387 */
1388 int
1389 puffs_vnop_reclaim(void *v)
1390 {
1391 struct vop_reclaim_args /* {
1392 const struct vnodeop_desc *a_desc;
1393 struct vnode *a_vp;
1394 } */ *ap = v;
1395 struct vnode *vp = ap->a_vp;
1396 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1397 struct puffs_node *pnode = vp->v_data;
1398 bool notifyserver = true;
1399
1400 /*
1401 * first things first: check if someone is trying to reclaim the
1402 * root vnode. do not allow that to travel to userspace.
1403 * Note that we don't need to take the lock similarly to
1404 * puffs_root(), since there is only one of us.
1405 */
1406 if (vp->v_vflag & VV_ROOT) {
1407 mutex_enter(&pmp->pmp_lock);
1408 KASSERT(pmp->pmp_root != NULL);
1409 pmp->pmp_root = NULL;
1410 mutex_exit(&pmp->pmp_lock);
1411 notifyserver = false;
1412 }
1413
1414 /*
1415 * purge info from kernel before issueing FAF, since we
1416 * don't really know when we'll get around to it after
1417 * that and someone might race us into node creation
1418 */
1419 mutex_enter(&pmp->pmp_lock);
1420 LIST_REMOVE(pnode, pn_hashent);
1421 if (PUFFS_USE_NAMECACHE(pmp))
1422 cache_purge(vp);
1423 mutex_exit(&pmp->pmp_lock);
1424
1425 if (notifyserver) {
1426 int nlookup = VPTOPP(vp)->pn_nlookup;
1427
1428 callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp), nlookup);
1429 }
1430
1431 if (PUFFS_USE_DOTDOTCACHE(pmp)) {
1432 if (__predict_true(VPTOPP(vp)->pn_parent != NULL))
1433 vrele(VPTOPP(vp)->pn_parent);
1434 else
1435 KASSERT(vp->v_vflag & VV_ROOT);
1436 }
1437
1438 puffs_putvnode(vp);
1439 vp->v_data = NULL;
1440
1441 return 0;
1442 }
1443
1444 #define CSIZE sizeof(**ap->a_cookies)
1445 int
1446 puffs_vnop_readdir(void *v)
1447 {
1448 struct vop_readdir_args /* {
1449 const struct vnodeop_desc *a_desc;
1450 struct vnode *a_vp;
1451 struct uio *a_uio;
1452 kauth_cred_t a_cred;
1453 int *a_eofflag;
1454 off_t **a_cookies;
1455 int *a_ncookies;
1456 } */ *ap = v;
1457 PUFFS_MSG_VARS(vn, readdir);
1458 struct vnode *vp = ap->a_vp;
1459 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1460 size_t argsize, tomove, cookiemem, cookiesmax;
1461 struct uio *uio = ap->a_uio;
1462 size_t howmuch, resid;
1463 int error;
1464
1465 /*
1466 * ok, so we need: resid + cookiemem = maxreq
1467 * => resid + cookiesize * (resid/minsize) = maxreq
1468 * => resid + cookiesize/minsize * resid = maxreq
1469 * => (cookiesize/minsize + 1) * resid = maxreq
1470 * => resid = maxreq / (cookiesize/minsize + 1)
1471 *
1472 * Since cookiesize <= minsize and we're not very big on floats,
1473 * we approximate that to be 1. Therefore:
1474 *
1475 * resid = maxreq / 2;
1476 *
1477 * Well, at least we didn't have to use differential equations
1478 * or the Gram-Schmidt process.
1479 *
1480 * (yes, I'm very afraid of this)
1481 */
1482 KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0));
1483
1484 if (ap->a_cookies) {
1485 KASSERT(ap->a_ncookies != NULL);
1486 if (pmp->pmp_args.pa_fhsize == 0)
1487 return EOPNOTSUPP;
1488 resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2;
1489 cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0);
1490 cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */
1491 } else {
1492 resid = PUFFS_TOMOVE(uio->uio_resid, pmp);
1493 cookiesmax = 0;
1494 cookiemem = 0;
1495 }
1496
1497 argsize = sizeof(struct puffs_vnmsg_readdir);
1498 tomove = resid + cookiemem;
1499 puffs_msgmem_alloc(argsize + tomove, &park_readdir,
1500 (void *)&readdir_msg, 1);
1501
1502 puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred);
1503 readdir_msg->pvnr_offset = uio->uio_offset;
1504 readdir_msg->pvnr_resid = resid;
1505 readdir_msg->pvnr_ncookies = cookiesmax;
1506 readdir_msg->pvnr_eofflag = 0;
1507 readdir_msg->pvnr_dentoff = cookiemem;
1508 puffs_msg_setinfo(park_readdir, PUFFSOP_VN,
1509 PUFFS_VN_READDIR, VPTOPNC(vp));
1510 puffs_msg_setdelta(park_readdir, tomove);
1511
1512 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error);
1513 error = checkerr(pmp, error, __func__);
1514 if (error)
1515 goto out;
1516
1517 /* userspace is cheating? */
1518 if (readdir_msg->pvnr_resid > resid) {
1519 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1520 "resid grew", VPTOPNC(vp));
1521 ERROUT(EPROTO);
1522 }
1523 if (readdir_msg->pvnr_ncookies > cookiesmax) {
1524 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG,
1525 "too many cookies", VPTOPNC(vp));
1526 ERROUT(EPROTO);
1527 }
1528
1529 /* check eof */
1530 if (readdir_msg->pvnr_eofflag)
1531 *ap->a_eofflag = 1;
1532
1533 /* bouncy-wouncy with the directory data */
1534 howmuch = resid - readdir_msg->pvnr_resid;
1535
1536 /* force eof if no data was returned (getcwd() needs this) */
1537 if (howmuch == 0) {
1538 *ap->a_eofflag = 1;
1539 goto out;
1540 }
1541
1542 error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio);
1543 if (error)
1544 goto out;
1545
1546 /* provide cookies to caller if so desired */
1547 if (ap->a_cookies) {
1548 KASSERT(curlwp != uvm.pagedaemon_lwp);
1549 *ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE,
1550 M_TEMP, M_WAITOK);
1551 *ap->a_ncookies = readdir_msg->pvnr_ncookies;
1552 memcpy(*ap->a_cookies, readdir_msg->pvnr_data,
1553 *ap->a_ncookies*CSIZE);
1554 }
1555
1556 /* next readdir starts here */
1557 uio->uio_offset = readdir_msg->pvnr_offset;
1558
1559 out:
1560 puffs_msgmem_release(park_readdir);
1561 return error;
1562 }
1563 #undef CSIZE
1564
1565 /*
1566 * poll works by consuming the bitmask in pn_revents. If there are
1567 * events available, poll returns immediately. If not, it issues a
1568 * poll to userspace, selrecords itself and returns with no available
1569 * events. When the file server returns, it executes puffs_parkdone_poll(),
1570 * where available events are added to the bitmask. selnotify() is
1571 * then also executed by that function causing us to enter here again
1572 * and hopefully find the missing bits (unless someone got them first,
1573 * in which case it starts all over again).
1574 */
1575 int
1576 puffs_vnop_poll(void *v)
1577 {
1578 struct vop_poll_args /* {
1579 const struct vnodeop_desc *a_desc;
1580 struct vnode *a_vp;
1581 int a_events;
1582 } */ *ap = v;
1583 PUFFS_MSG_VARS(vn, poll);
1584 struct vnode *vp = ap->a_vp;
1585 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1586 struct puffs_node *pn = vp->v_data;
1587 int events;
1588
1589 if (EXISTSOP(pmp, POLL)) {
1590 mutex_enter(&pn->pn_mtx);
1591 events = pn->pn_revents & ap->a_events;
1592 if (events & ap->a_events) {
1593 pn->pn_revents &= ~ap->a_events;
1594 mutex_exit(&pn->pn_mtx);
1595
1596 return events;
1597 } else {
1598 puffs_referencenode(pn);
1599 mutex_exit(&pn->pn_mtx);
1600
1601 PUFFS_MSG_ALLOC(vn, poll);
1602 poll_msg->pvnr_events = ap->a_events;
1603 puffs_msg_setinfo(park_poll, PUFFSOP_VN,
1604 PUFFS_VN_POLL, VPTOPNC(vp));
1605 puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn);
1606 selrecord(curlwp, &pn->pn_sel);
1607
1608 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_poll,
1609 vp->v_data, NULL);
1610 PUFFS_MSG_RELEASE(poll);
1611
1612 return 0;
1613 }
1614 } else {
1615 return genfs_poll(v);
1616 }
1617 }
1618
1619 static int
1620 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait)
1621 {
1622 struct puffs_node *pn = VPTOPP(vp);
1623 struct vattr va;
1624 int pflags, error;
1625
1626 /* flush out information from our metacache, see vop_setattr */
1627 if (pn->pn_stat & PNODE_METACACHE_MASK
1628 && (pn->pn_stat & PNODE_DYING) == 0) {
1629 vattr_null(&va);
1630 error = dosetattr(vp, &va, FSCRED,
1631 SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC));
1632 if (error)
1633 return error;
1634 }
1635
1636 /*
1637 * flush pages to avoid being overly dirty
1638 */
1639 pflags = PGO_CLEANIT;
1640 if (wait)
1641 pflags |= PGO_SYNCIO;
1642
1643 mutex_enter(vp->v_interlock);
1644 return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags);
1645 }
1646
1647 int
1648 puffs_vnop_fsync(void *v)
1649 {
1650 struct vop_fsync_args /* {
1651 const struct vnodeop_desc *a_desc;
1652 struct vnode *a_vp;
1653 kauth_cred_t a_cred;
1654 int a_flags;
1655 off_t a_offlo;
1656 off_t a_offhi;
1657 } */ *ap = v;
1658 PUFFS_MSG_VARS(vn, fsync);
1659 struct vnode *vp;
1660 struct puffs_node *pn;
1661 struct puffs_mount *pmp;
1662 int error, dofaf;
1663
1664 vp = ap->a_vp;
1665 KASSERT(vp != NULL);
1666 pn = VPTOPP(vp);
1667 KASSERT(pn != NULL);
1668 pmp = MPTOPUFFSMP(vp->v_mount);
1669 if (ap->a_flags & FSYNC_WAIT) {
1670 mutex_enter(&pn->pn_sizemtx);
1671 } else {
1672 if (mutex_tryenter(&pn->pn_sizemtx) == 0)
1673 return EDEADLK;
1674 }
1675
1676 error = flushvncache(vp, ap->a_offlo, ap->a_offhi,
1677 (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT);
1678 if (error)
1679 goto out;
1680
1681 /*
1682 * HELLO! We exit already here if the user server does not
1683 * support fsync OR if we should call fsync for a node which
1684 * has references neither in the kernel or the fs server.
1685 * Otherwise we continue to issue fsync() forward.
1686 */
1687 error = 0;
1688 if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING))
1689 goto out;
1690
1691 dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY;
1692 /*
1693 * We abuse VXLOCK to mean "vnode is going to die", so we issue
1694 * only FAFs for those. Otherwise there's a danger of deadlock,
1695 * since the execution context here might be the user server
1696 * doing some operation on another fs, which in turn caused a
1697 * vnode to be reclaimed from the freelist for this fs.
1698 */
1699 if (dofaf == 0) {
1700 mutex_enter(vp->v_interlock);
1701 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
1702 dofaf = 1;
1703 mutex_exit(vp->v_interlock);
1704 }
1705
1706 PUFFS_MSG_ALLOC(vn, fsync);
1707 if (dofaf)
1708 puffs_msg_setfaf(park_fsync);
1709
1710 puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred);
1711 fsync_msg->pvnr_flags = ap->a_flags;
1712 fsync_msg->pvnr_offlo = ap->a_offlo;
1713 fsync_msg->pvnr_offhi = ap->a_offhi;
1714 puffs_msg_setinfo(park_fsync, PUFFSOP_VN,
1715 PUFFS_VN_FSYNC, VPTOPNC(vp));
1716
1717 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error);
1718 PUFFS_MSG_RELEASE(fsync);
1719
1720 error = checkerr(pmp, error, __func__);
1721
1722 out:
1723 mutex_exit(&pn->pn_sizemtx);
1724 return error;
1725 }
1726
1727 int
1728 puffs_vnop_seek(void *v)
1729 {
1730 struct vop_seek_args /* {
1731 const struct vnodeop_desc *a_desc;
1732 struct vnode *a_vp;
1733 off_t a_oldoff;
1734 off_t a_newoff;
1735 kauth_cred_t a_cred;
1736 } */ *ap = v;
1737 PUFFS_MSG_VARS(vn, seek);
1738 struct vnode *vp = ap->a_vp;
1739 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1740 int error;
1741
1742 PUFFS_MSG_ALLOC(vn, seek);
1743 seek_msg->pvnr_oldoff = ap->a_oldoff;
1744 seek_msg->pvnr_newoff = ap->a_newoff;
1745 puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred);
1746 puffs_msg_setinfo(park_seek, PUFFSOP_VN,
1747 PUFFS_VN_SEEK, VPTOPNC(vp));
1748
1749 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error);
1750 PUFFS_MSG_RELEASE(seek);
1751 return checkerr(pmp, error, __func__);
1752 }
1753
1754 static int
1755 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1756 struct componentname *cnp)
1757 {
1758 PUFFS_MSG_VARS(vn, remove);
1759 int error;
1760
1761 PUFFS_MSG_ALLOC(vn, remove);
1762 remove_msg->pvnr_cookie_targ = ck;
1763 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1764 cnp, PUFFS_USE_FULLPNBUF(pmp));
1765 puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck);
1766
1767 PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error);
1768 PUFFS_MSG_RELEASE(remove);
1769
1770 return checkerr(pmp, error, __func__);
1771 }
1772
1773 /*
1774 * XXX: can't use callremove now because can't catch setbacks with
1775 * it due to lack of a pnode argument.
1776 */
1777 int
1778 puffs_vnop_remove(void *v)
1779 {
1780 struct vop_remove_args /* {
1781 const struct vnodeop_desc *a_desc;
1782 struct vnode *a_dvp;
1783 struct vnode *a_vp;
1784 struct componentname *a_cnp;
1785 } */ *ap = v;
1786 PUFFS_MSG_VARS(vn, remove);
1787 struct vnode *dvp = ap->a_dvp;
1788 struct vnode *vp = ap->a_vp;
1789 struct puffs_node *dpn = VPTOPP(dvp);
1790 struct puffs_node *pn = VPTOPP(vp);
1791 struct componentname *cnp = ap->a_cnp;
1792 struct mount *mp = dvp->v_mount;
1793 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1794 int error;
1795
1796 PUFFS_MSG_ALLOC(vn, remove);
1797 remove_msg->pvnr_cookie_targ = VPTOPNC(vp);
1798 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred,
1799 cnp, PUFFS_USE_FULLPNBUF(pmp));
1800 puffs_msg_setinfo(park_remove, PUFFSOP_VN,
1801 PUFFS_VN_REMOVE, VPTOPNC(dvp));
1802
1803 puffs_msg_enqueue(pmp, park_remove);
1804 REFPN_AND_UNLOCKVP(dvp, dpn);
1805 if (dvp == vp)
1806 REFPN(pn);
1807 else
1808 REFPN_AND_UNLOCKVP(vp, pn);
1809 error = puffs_msg_wait2(pmp, park_remove, dpn, pn);
1810
1811 PUFFS_MSG_RELEASE(remove);
1812
1813 RELEPN_AND_VP(dvp, dpn);
1814 RELEPN_AND_VP(vp, pn);
1815
1816 error = checkerr(pmp, error, __func__);
1817 return error;
1818 }
1819
1820 int
1821 puffs_vnop_mkdir(void *v)
1822 {
1823 struct vop_mkdir_v3_args /* {
1824 const struct vnodeop_desc *a_desc;
1825 struct vnode *a_dvp;
1826 struct vnode **a_vpp;
1827 struct componentname *a_cnp;
1828 struct vattr *a_vap;
1829 } */ *ap = v;
1830 PUFFS_MSG_VARS(vn, mkdir);
1831 struct vnode *dvp = ap->a_dvp;
1832 struct puffs_node *dpn = VPTOPP(dvp);
1833 struct componentname *cnp = ap->a_cnp;
1834 struct mount *mp = dvp->v_mount;
1835 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1836 int error;
1837
1838 PUFFS_MSG_ALLOC(vn, mkdir);
1839 puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred,
1840 cnp, PUFFS_USE_FULLPNBUF(pmp));
1841 mkdir_msg->pvnr_va = *ap->a_vap;
1842 puffs_msg_setinfo(park_mkdir, PUFFSOP_VN,
1843 PUFFS_VN_MKDIR, VPTOPNC(dvp));
1844
1845 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error);
1846
1847 error = checkerr(pmp, error, __func__);
1848 if (error)
1849 goto out;
1850
1851 error = puffs_newnode(mp, dvp, ap->a_vpp,
1852 mkdir_msg->pvnr_newnode, cnp, VDIR, 0);
1853 if (error) {
1854 puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie,
1855 mkdir_msg->pvnr_newnode, cnp);
1856 goto out;
1857 }
1858
1859 if (PUFFS_USE_FS_TTL(pmp)) {
1860 struct timespec *va_ttl = &mkdir_msg->pvnr_va_ttl;
1861 struct timespec *cn_ttl = &mkdir_msg->pvnr_cn_ttl;
1862 struct vattr *rvap = &mkdir_msg->pvnr_va;
1863
1864 update_va(*ap->a_vpp, NULL, rvap,
1865 va_ttl, cn_ttl, SETATTR_CHSIZE);
1866 }
1867
1868 VPTOPP(*ap->a_vpp)->pn_nlookup++;
1869
1870 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
1871 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
1872 update_parent(*ap->a_vpp, dvp);
1873
1874 out:
1875 PUFFS_MSG_RELEASE(mkdir);
1876 return error;
1877 }
1878
1879 static int
1880 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1881 struct componentname *cnp)
1882 {
1883 PUFFS_MSG_VARS(vn, rmdir);
1884 int error;
1885
1886 PUFFS_MSG_ALLOC(vn, rmdir);
1887 rmdir_msg->pvnr_cookie_targ = ck;
1888 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1889 cnp, PUFFS_USE_FULLPNBUF(pmp));
1890 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1891
1892 PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1893 PUFFS_MSG_RELEASE(rmdir);
1894
1895 return checkerr(pmp, error, __func__);
1896 }
1897
1898 int
1899 puffs_vnop_rmdir(void *v)
1900 {
1901 struct vop_rmdir_args /* {
1902 const struct vnodeop_desc *a_desc;
1903 struct vnode *a_dvp;
1904 struct vnode *a_vp;
1905 struct componentname *a_cnp;
1906 } */ *ap = v;
1907 PUFFS_MSG_VARS(vn, rmdir);
1908 struct vnode *dvp = ap->a_dvp;
1909 struct vnode *vp = ap->a_vp;
1910 struct puffs_node *dpn = VPTOPP(dvp);
1911 struct puffs_node *pn = VPTOPP(vp);
1912 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1913 struct componentname *cnp = ap->a_cnp;
1914 int error;
1915
1916 PUFFS_MSG_ALLOC(vn, rmdir);
1917 rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1918 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1919 cnp, PUFFS_USE_FULLPNBUF(pmp));
1920 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1921 PUFFS_VN_RMDIR, VPTOPNC(dvp));
1922
1923 puffs_msg_enqueue(pmp, park_rmdir);
1924 REFPN_AND_UNLOCKVP(dvp, dpn);
1925 REFPN_AND_UNLOCKVP(vp, pn);
1926 error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
1927
1928 PUFFS_MSG_RELEASE(rmdir);
1929
1930 /* XXX: some call cache_purge() *for both vnodes* here, investigate */
1931 RELEPN_AND_VP(dvp, dpn);
1932 RELEPN_AND_VP(vp, pn);
1933
1934 return error;
1935 }
1936
1937 int
1938 puffs_vnop_link(void *v)
1939 {
1940 struct vop_link_args /* {
1941 const struct vnodeop_desc *a_desc;
1942 struct vnode *a_dvp;
1943 struct vnode *a_vp;
1944 struct componentname *a_cnp;
1945 } */ *ap = v;
1946 PUFFS_MSG_VARS(vn, link);
1947 struct vnode *dvp = ap->a_dvp;
1948 struct vnode *vp = ap->a_vp;
1949 struct puffs_node *dpn = VPTOPP(dvp);
1950 struct puffs_node *pn = VPTOPP(vp);
1951 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1952 struct componentname *cnp = ap->a_cnp;
1953 int error;
1954
1955 PUFFS_MSG_ALLOC(vn, link);
1956 link_msg->pvnr_cookie_targ = VPTOPNC(vp);
1957 puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
1958 cnp, PUFFS_USE_FULLPNBUF(pmp));
1959 puffs_msg_setinfo(park_link, PUFFSOP_VN,
1960 PUFFS_VN_LINK, VPTOPNC(dvp));
1961
1962 puffs_msg_enqueue(pmp, park_link);
1963 REFPN_AND_UNLOCKVP(dvp, dpn);
1964 REFPN(pn);
1965 error = puffs_msg_wait2(pmp, park_link, dpn, pn);
1966
1967 PUFFS_MSG_RELEASE(link);
1968
1969 error = checkerr(pmp, error, __func__);
1970
1971 /*
1972 * XXX: stay in touch with the cache. I don't like this, but
1973 * don't have a better solution either. See also puffs_rename().
1974 */
1975 if (error == 0)
1976 puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
1977
1978 RELEPN_AND_VP(dvp, dpn);
1979 puffs_releasenode(pn);
1980
1981 return error;
1982 }
1983
1984 int
1985 puffs_vnop_symlink(void *v)
1986 {
1987 struct vop_symlink_v3_args /* {
1988 const struct vnodeop_desc *a_desc;
1989 struct vnode *a_dvp;
1990 struct vnode **a_vpp;
1991 struct componentname *a_cnp;
1992 struct vattr *a_vap;
1993 char *a_target;
1994 } */ *ap = v;
1995 PUFFS_MSG_VARS(vn, symlink);
1996 struct vnode *dvp = ap->a_dvp;
1997 struct puffs_node *dpn = VPTOPP(dvp);
1998 struct mount *mp = dvp->v_mount;
1999 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2000 struct componentname *cnp = ap->a_cnp;
2001 int error;
2002
2003 *ap->a_vpp = NULL;
2004
2005 PUFFS_MSG_ALLOC(vn, symlink);
2006 puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
2007 cnp, PUFFS_USE_FULLPNBUF(pmp));
2008 symlink_msg->pvnr_va = *ap->a_vap;
2009 (void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
2010 sizeof(symlink_msg->pvnr_link));
2011 puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
2012 PUFFS_VN_SYMLINK, VPTOPNC(dvp));
2013
2014 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
2015
2016 error = checkerr(pmp, error, __func__);
2017 if (error)
2018 goto out;
2019
2020 error = puffs_newnode(mp, dvp, ap->a_vpp,
2021 symlink_msg->pvnr_newnode, cnp, VLNK, 0);
2022 if (error) {
2023 puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
2024 symlink_msg->pvnr_newnode, cnp);
2025 goto out;
2026 }
2027
2028 if (PUFFS_USE_FS_TTL(pmp)) {
2029 struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl;
2030 struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl;
2031 struct vattr *rvap = &symlink_msg->pvnr_va;
2032
2033 update_va(*ap->a_vpp, NULL, rvap,
2034 va_ttl, cn_ttl, SETATTR_CHSIZE);
2035 }
2036
2037 VPTOPP(*ap->a_vpp)->pn_nlookup++;
2038
2039 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2040 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
2041 update_parent(*ap->a_vpp, dvp);
2042
2043 out:
2044 PUFFS_MSG_RELEASE(symlink);
2045
2046 return error;
2047 }
2048
2049 int
2050 puffs_vnop_readlink(void *v)
2051 {
2052 struct vop_readlink_args /* {
2053 const struct vnodeop_desc *a_desc;
2054 struct vnode *a_vp;
2055 struct uio *a_uio;
2056 kauth_cred_t a_cred;
2057 } */ *ap = v;
2058 PUFFS_MSG_VARS(vn, readlink);
2059 struct vnode *vp = ap->a_vp;
2060 struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2061 size_t linklen;
2062 int error;
2063
2064 PUFFS_MSG_ALLOC(vn, readlink);
2065 puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
2066 linklen = sizeof(readlink_msg->pvnr_link);
2067 readlink_msg->pvnr_linklen = linklen;
2068 puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
2069 PUFFS_VN_READLINK, VPTOPNC(vp));
2070
2071 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
2072 error = checkerr(pmp, error, __func__);
2073 if (error)
2074 goto out;
2075
2076 /* bad bad user file server */
2077 if (readlink_msg->pvnr_linklen > linklen) {
2078 puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
2079 "linklen too big", VPTOPNC(ap->a_vp));
2080 error = EPROTO;
2081 goto out;
2082 }
2083
2084 error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
2085 ap->a_uio);
2086 out:
2087 PUFFS_MSG_RELEASE(readlink);
2088 return error;
2089 }
2090
2091 int
2092 puffs_vnop_rename(void *v)
2093 {
2094 struct vop_rename_args /* {
2095 const struct vnodeop_desc *a_desc;
2096 struct vnode *a_fdvp;
2097 struct vnode *a_fvp;
2098 struct componentname *a_fcnp;
2099 struct vnode *a_tdvp;
2100 struct vnode *a_tvp;
2101 struct componentname *a_tcnp;
2102 } */ *ap = v;
2103 PUFFS_MSG_VARS(vn, rename);
2104 struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
2105 struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
2106 struct puffs_node *fpn = ap->a_fvp->v_data;
2107 struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
2108 int error;
2109 bool doabort = true;
2110
2111 if ((fvp->v_mount != tdvp->v_mount) ||
2112 (tvp && (fvp->v_mount != tvp->v_mount))) {
2113 ERROUT(EXDEV);
2114 }
2115
2116 PUFFS_MSG_ALLOC(vn, rename);
2117 rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
2118 rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
2119 if (tvp)
2120 rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
2121 else
2122 rename_msg->pvnr_cookie_targ = NULL;
2123 puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
2124 ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
2125 puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
2126 ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
2127 puffs_msg_setinfo(park_rename, PUFFSOP_VN,
2128 PUFFS_VN_RENAME, VPTOPNC(fdvp));
2129
2130 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
2131 doabort = false;
2132 PUFFS_MSG_RELEASE(rename);
2133 error = checkerr(pmp, error, __func__);
2134
2135 /*
2136 * XXX: stay in touch with the cache. I don't like this, but
2137 * don't have a better solution either. See also puffs_link().
2138 */
2139 if (error == 0) {
2140 puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
2141
2142 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2143 (VPTOPP(fvp)->pn_parent != tdvp))
2144 update_parent(fvp, tdvp);
2145 }
2146
2147
2148 out:
2149 if (doabort)
2150 VOP_ABORTOP(tdvp, ap->a_tcnp);
2151 if (tvp != NULL)
2152 vput(tvp);
2153 if (tdvp == tvp)
2154 vrele(tdvp);
2155 else
2156 vput(tdvp);
2157
2158 if (doabort)
2159 VOP_ABORTOP(fdvp, ap->a_fcnp);
2160 vrele(fdvp);
2161 vrele(fvp);
2162
2163 return error;
2164 }
2165
2166 #define RWARGS(cont, iofl, move, offset, creds) \
2167 (cont)->pvnr_ioflag = (iofl); \
2168 (cont)->pvnr_resid = (move); \
2169 (cont)->pvnr_offset = (offset); \
2170 puffs_credcvt(&(cont)->pvnr_cred, creds)
2171
2172 int
2173 puffs_vnop_read(void *v)
2174 {
2175 struct vop_read_args /* {
2176 const struct vnodeop_desc *a_desc;
2177 struct vnode *a_vp;
2178 struct uio *a_uio;
2179 int a_ioflag;
2180 kauth_cred_t a_cred;
2181 } */ *ap = v;
2182 PUFFS_MSG_VARS(vn, read);
2183 struct vnode *vp = ap->a_vp;
2184 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2185 struct uio *uio = ap->a_uio;
2186 size_t tomove, argsize;
2187 vsize_t bytelen;
2188 int error;
2189
2190 read_msg = NULL;
2191 error = 0;
2192
2193 /* std sanity */
2194 if (uio->uio_resid == 0)
2195 return 0;
2196 if (uio->uio_offset < 0)
2197 return EINVAL;
2198
2199 if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
2200 const int advice = IO_ADV_DECODE(ap->a_ioflag);
2201
2202 while (uio->uio_resid > 0) {
2203 if (vp->v_size <= uio->uio_offset) {
2204 break;
2205 }
2206 bytelen = MIN(uio->uio_resid,
2207 vp->v_size - uio->uio_offset);
2208 if (bytelen == 0)
2209 break;
2210
2211 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
2212 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
2213 if (error)
2214 break;
2215 }
2216
2217 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
2218 puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
2219 } else {
2220 /*
2221 * in case it's not a regular file or we're operating
2222 * uncached, do read in the old-fashioned style,
2223 * i.e. explicit read operations
2224 */
2225
2226 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2227 argsize = sizeof(struct puffs_vnmsg_read);
2228 puffs_msgmem_alloc(argsize + tomove, &park_read,
2229 (void *)&read_msg, 1);
2230
2231 error = 0;
2232 while (uio->uio_resid > 0) {
2233 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2234 memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
2235 RWARGS(read_msg, ap->a_ioflag, tomove,
2236 uio->uio_offset, ap->a_cred);
2237 puffs_msg_setinfo(park_read, PUFFSOP_VN,
2238 PUFFS_VN_READ, VPTOPNC(vp));
2239 puffs_msg_setdelta(park_read, tomove);
2240
2241 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
2242 NULL, error);
2243 error = checkerr(pmp, error, __func__);
2244 if (error)
2245 break;
2246
2247 if (read_msg->pvnr_resid > tomove) {
2248 puffs_senderr(pmp, PUFFS_ERR_READ,
2249 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2250 error = EPROTO;
2251 break;
2252 }
2253
2254 error = uiomove(read_msg->pvnr_data,
2255 tomove - read_msg->pvnr_resid, uio);
2256
2257 /*
2258 * in case the file is out of juice, resid from
2259 * userspace is != 0. and the error-case is
2260 * quite obvious
2261 */
2262 if (error || read_msg->pvnr_resid)
2263 break;
2264 }
2265
2266 puffs_msgmem_release(park_read);
2267 }
2268
2269 return error;
2270 }
2271
2272 /*
2273 * XXX: in case of a failure, this leaves uio in a bad state.
2274 * We could theoretically copy the uio and iovecs and "replay"
2275 * them the right amount after the userspace trip, but don't
2276 * bother for now.
2277 */
2278 int
2279 puffs_vnop_write(void *v)
2280 {
2281 struct vop_write_args /* {
2282 const struct vnodeop_desc *a_desc;
2283 struct vnode *a_vp;
2284 struct uio *a_uio;
2285 int a_ioflag;
2286 kauth_cred_t a_cred;
2287 } */ *ap = v;
2288 PUFFS_MSG_VARS(vn, write);
2289 struct vnode *vp = ap->a_vp;
2290 struct puffs_node *pn = VPTOPP(vp);
2291 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2292 struct uio *uio = ap->a_uio;
2293 size_t tomove, argsize;
2294 off_t oldoff, newoff, origoff;
2295 vsize_t bytelen;
2296 int error, uflags;
2297 int ubcflags;
2298
2299 error = uflags = 0;
2300 write_msg = NULL;
2301
2302 mutex_enter(&pn->pn_sizemtx);
2303
2304 if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
2305 ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp);
2306
2307 /*
2308 * userspace *should* be allowed to control this,
2309 * but with UBC it's a bit unclear how to handle it
2310 */
2311 if (ap->a_ioflag & IO_APPEND)
2312 uio->uio_offset = vp->v_size;
2313
2314 origoff = uio->uio_offset;
2315 while (uio->uio_resid > 0) {
2316 if (vp->v_mount->mnt_flag & MNT_RELATIME)
2317 uflags |= PUFFS_UPDATEATIME;
2318 uflags |= PUFFS_UPDATECTIME;
2319 uflags |= PUFFS_UPDATEMTIME;
2320 oldoff = uio->uio_offset;
2321 bytelen = uio->uio_resid;
2322
2323 newoff = oldoff + bytelen;
2324 if (vp->v_size < newoff) {
2325 uvm_vnp_setwritesize(vp, newoff);
2326 }
2327 error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
2328 UVM_ADV_RANDOM, ubcflags);
2329
2330 /*
2331 * In case of a ubc_uiomove() error,
2332 * opt to not extend the file at all and
2333 * return an error. Otherwise, if we attempt
2334 * to clear the memory we couldn't fault to,
2335 * we might generate a kernel page fault.
2336 */
2337 if (vp->v_size < newoff) {
2338 if (error == 0) {
2339 uflags |= PUFFS_UPDATESIZE;
2340 uvm_vnp_setsize(vp, newoff);
2341 } else {
2342 uvm_vnp_setwritesize(vp, vp->v_size);
2343 }
2344 }
2345 if (error)
2346 break;
2347
2348 /*
2349 * If we're writing large files, flush to file server
2350 * every 64k. Otherwise we can very easily exhaust
2351 * kernel and user memory, as the file server cannot
2352 * really keep up with our writing speed.
2353 *
2354 * Note: this does *NOT* honor MNT_ASYNC, because
2355 * that gives userland too much say in the kernel.
2356 */
2357 if (oldoff >> 16 != uio->uio_offset >> 16) {
2358 mutex_enter(vp->v_interlock);
2359 error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
2360 uio->uio_offset & ~0xffff,
2361 PGO_CLEANIT | PGO_SYNCIO);
2362 if (error)
2363 break;
2364 }
2365 }
2366
2367 /* synchronous I/O? */
2368 if (error == 0 && ap->a_ioflag & IO_SYNC) {
2369 mutex_enter(vp->v_interlock);
2370 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2371 round_page(uio->uio_offset),
2372 PGO_CLEANIT | PGO_SYNCIO);
2373
2374 /* write through page cache? */
2375 } else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
2376 mutex_enter(vp->v_interlock);
2377 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2378 round_page(uio->uio_offset), PGO_CLEANIT);
2379 }
2380
2381 puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
2382 } else {
2383 /* tomove is non-increasing */
2384 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2385 argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2386 puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2387
2388 while (uio->uio_resid > 0) {
2389 /* move data to buffer */
2390 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2391 memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2392 RWARGS(write_msg, ap->a_ioflag, tomove,
2393 uio->uio_offset, ap->a_cred);
2394 error = uiomove(write_msg->pvnr_data, tomove, uio);
2395 if (error)
2396 break;
2397
2398 /* move buffer to userspace */
2399 puffs_msg_setinfo(park_write, PUFFSOP_VN,
2400 PUFFS_VN_WRITE, VPTOPNC(vp));
2401 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2402 NULL, error);
2403 error = checkerr(pmp, error, __func__);
2404 if (error)
2405 break;
2406
2407 if (write_msg->pvnr_resid > tomove) {
2408 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2409 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2410 error = EPROTO;
2411 break;
2412 }
2413
2414 /* adjust file size */
2415 if (vp->v_size < uio->uio_offset)
2416 uvm_vnp_setsize(vp, uio->uio_offset);
2417
2418 /* didn't move everything? bad userspace. bail */
2419 if (write_msg->pvnr_resid != 0) {
2420 error = EIO;
2421 break;
2422 }
2423 }
2424 puffs_msgmem_release(park_write);
2425 }
2426
2427 mutex_exit(&pn->pn_sizemtx);
2428 return error;
2429 }
2430
2431 int
2432 puffs_vnop_print(void *v)
2433 {
2434 struct vop_print_args /* {
2435 struct vnode *a_vp;
2436 } */ *ap = v;
2437 PUFFS_MSG_VARS(vn, print);
2438 struct vnode *vp = ap->a_vp;
2439 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2440 struct puffs_node *pn = vp->v_data;
2441
2442 /* kernel portion */
2443 printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2444 "\tuserspace cookie: %p", vp, pn, pn->pn_cookie);
2445 if (vp->v_type == VFIFO)
2446 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
2447 printf("\n");
2448
2449 /* userspace portion */
2450 if (EXISTSOP(pmp, PRINT)) {
2451 PUFFS_MSG_ALLOC(vn, print);
2452 puffs_msg_setinfo(park_print, PUFFSOP_VN,
2453 PUFFS_VN_PRINT, VPTOPNC(vp));
2454 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_print, vp->v_data,
2455 NULL);
2456 PUFFS_MSG_RELEASE(print);
2457 }
2458
2459 return 0;
2460 }
2461
2462 int
2463 puffs_vnop_pathconf(void *v)
2464 {
2465 struct vop_pathconf_args /* {
2466 const struct vnodeop_desc *a_desc;
2467 struct vnode *a_vp;
2468 int a_name;
2469 register_t *a_retval;
2470 } */ *ap = v;
2471 PUFFS_MSG_VARS(vn, pathconf);
2472 struct vnode *vp = ap->a_vp;
2473 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2474 int error;
2475
2476 PUFFS_MSG_ALLOC(vn, pathconf);
2477 pathconf_msg->pvnr_name = ap->a_name;
2478 puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2479 PUFFS_VN_PATHCONF, VPTOPNC(vp));
2480 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2481 error = checkerr(pmp, error, __func__);
2482 if (!error)
2483 *ap->a_retval = pathconf_msg->pvnr_retval;
2484 PUFFS_MSG_RELEASE(pathconf);
2485
2486 return error;
2487 }
2488
2489 int
2490 puffs_vnop_advlock(void *v)
2491 {
2492 struct vop_advlock_args /* {
2493 const struct vnodeop_desc *a_desc;
2494 struct vnode *a_vp;
2495 void *a_id;
2496 int a_op;
2497 struct flock *a_fl;
2498 int a_flags;
2499 } */ *ap = v;
2500 PUFFS_MSG_VARS(vn, advlock);
2501 struct vnode *vp = ap->a_vp;
2502 struct puffs_node *pn = VPTOPP(vp);
2503 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2504 int error;
2505
2506 if (!EXISTSOP(pmp, ADVLOCK))
2507 return lf_advlock(ap, &pn->pn_lockf, vp->v_size);
2508
2509 PUFFS_MSG_ALLOC(vn, advlock);
2510 (void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl,
2511 sizeof(advlock_msg->pvnr_fl));
2512 advlock_msg->pvnr_id = ap->a_id;
2513 advlock_msg->pvnr_op = ap->a_op;
2514 advlock_msg->pvnr_flags = ap->a_flags;
2515 puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2516 PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2517 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2518 error = checkerr(pmp, error, __func__);
2519 PUFFS_MSG_RELEASE(advlock);
2520
2521 return error;
2522 }
2523
2524 int
2525 puffs_vnop_abortop(void *v)
2526 {
2527 struct vop_abortop_args /* {
2528 struct vnode *a_dvp;
2529 struct componentname *a_cnp;
2530 }; */ *ap = v;
2531 PUFFS_MSG_VARS(vn, abortop);
2532 struct vnode *dvp = ap->a_dvp;
2533 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2534 struct componentname *cnp = ap->a_cnp;
2535
2536 if (EXISTSOP(pmp, ABORTOP)) {
2537 PUFFS_MSG_ALLOC(vn, abortop);
2538 puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2539 cnp, PUFFS_USE_FULLPNBUF(pmp));
2540 puffs_msg_setfaf(park_abortop);
2541 puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2542 PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2543
2544 puffs_msg_enqueue(pmp, park_abortop);
2545 PUFFS_MSG_RELEASE(abortop);
2546 }
2547
2548 return genfs_abortop(v);
2549 }
2550
2551 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2552
2553 /*
2554 * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2555 */
2556 int
2557 puffs_vnop_strategy(void *v)
2558 {
2559 struct vop_strategy_args /* {
2560 const struct vnodeop_desc *a_desc;
2561 struct vnode *a_vp;
2562 struct buf *a_bp;
2563 } */ *ap = v;
2564 PUFFS_MSG_VARS(vn, rw);
2565 struct vnode *vp = ap->a_vp;
2566 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2567 struct puffs_node *pn;
2568 struct buf *bp;
2569 size_t argsize;
2570 size_t tomove, moved;
2571 int error, dofaf, cansleep, dobiodone;
2572
2573 pmp = MPTOPUFFSMP(vp->v_mount);
2574 bp = ap->a_bp;
2575 error = 0;
2576 dofaf = 0;
2577 cansleep = 0;
2578 pn = VPTOPP(vp);
2579 park_rw = NULL; /* explicit */
2580 dobiodone = 1;
2581
2582 if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2583 || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2584 ERROUT(EOPNOTSUPP);
2585
2586 /*
2587 * Short-circuit optimization: don't flush buffer in between
2588 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2589 */
2590 if (pn->pn_stat & PNODE_DYING) {
2591 KASSERT(BUF_ISWRITE(bp));
2592 bp->b_resid = 0;
2593 goto out;
2594 }
2595
2596 #ifdef DIAGNOSTIC
2597 if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2598 panic("puffs_strategy: wildly inappropriate buf bcount %d",
2599 bp->b_bcount);
2600 #endif
2601
2602 /*
2603 * See explanation for the necessity of a FAF in puffs_fsync.
2604 *
2605 * Also, do FAF in case we're suspending.
2606 * See puffs_vfsops.c:pageflush()
2607 */
2608 if (BUF_ISWRITE(bp)) {
2609 mutex_enter(vp->v_interlock);
2610 if (vdead_check(vp, VDEAD_NOWAIT) != 0)
2611 dofaf = 1;
2612 if (pn->pn_stat & PNODE_FAF)
2613 dofaf = 1;
2614 mutex_exit(vp->v_interlock);
2615 }
2616
2617 cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1;
2618
2619 KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp));
2620
2621 /* allocate transport structure */
2622 tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2623 argsize = sizeof(struct puffs_vnmsg_rw);
2624 error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2625 (void *)&rw_msg, cansleep);
2626 if (error)
2627 goto out;
2628 RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2629
2630 /* 2x2 cases: read/write, faf/nofaf */
2631 if (BUF_ISREAD(bp)) {
2632 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2633 PUFFS_VN_READ, VPTOPNC(vp));
2634 puffs_msg_setdelta(park_rw, tomove);
2635 if (BIOASYNC(bp)) {
2636 puffs_msg_setcall(park_rw,
2637 puffs_parkdone_asyncbioread, bp);
2638 puffs_msg_enqueue(pmp, park_rw);
2639 dobiodone = 0;
2640 } else {
2641 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2642 NULL, error);
2643 error = checkerr(pmp, error, __func__);
2644 if (error)
2645 goto out;
2646
2647 if (rw_msg->pvnr_resid > tomove) {
2648 puffs_senderr(pmp, PUFFS_ERR_READ,
2649 E2BIG, "resid grew", VPTOPNC(vp));
2650 ERROUT(EPROTO);
2651 }
2652
2653 moved = tomove - rw_msg->pvnr_resid;
2654
2655 (void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2656 bp->b_resid = bp->b_bcount - moved;
2657 }
2658 } else {
2659 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2660 PUFFS_VN_WRITE, VPTOPNC(vp));
2661 /*
2662 * make pages read-only before we write them if we want
2663 * write caching info
2664 */
2665 if (PUFFS_WCACHEINFO(pmp)) {
2666 struct uvm_object *uobj = &vp->v_uobj;
2667 int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2668 struct vm_page *vmp;
2669 int i;
2670
2671 for (i = 0; i < npages; i++) {
2672 vmp= uvm_pageratop((vaddr_t)bp->b_data
2673 + (i << PAGE_SHIFT));
2674 DPRINTF(("puffs_strategy: write-protecting "
2675 "vp %p page %p, offset %" PRId64"\n",
2676 vp, vmp, vmp->offset));
2677 mutex_enter(uobj->vmobjlock);
2678 vmp->flags |= PG_RDONLY;
2679 pmap_page_protect(vmp, VM_PROT_READ);
2680 mutex_exit(uobj->vmobjlock);
2681 }
2682 }
2683
2684 (void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2685 if (dofaf) {
2686 puffs_msg_setfaf(park_rw);
2687 } else if (BIOASYNC(bp)) {
2688 puffs_msg_setcall(park_rw,
2689 puffs_parkdone_asyncbiowrite, bp);
2690 dobiodone = 0;
2691 }
2692
2693 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2694
2695 if (dobiodone == 0)
2696 goto out;
2697
2698 /*
2699 * XXXXXXXX: wrong, but kernel can't survive strategy
2700 * failure currently. Here, have one more X: X.
2701 */
2702 if (error != ENOMEM)
2703 error = 0;
2704
2705 error = checkerr(pmp, error, __func__);
2706 if (error)
2707 goto out;
2708
2709 if (rw_msg->pvnr_resid > tomove) {
2710 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2711 E2BIG, "resid grew", VPTOPNC(vp));
2712 ERROUT(EPROTO);
2713 }
2714
2715 /*
2716 * FAF moved everything. Frankly, we don't
2717 * really have a choice.
2718 */
2719 if (dofaf && error == 0)
2720 moved = tomove;
2721 else
2722 moved = tomove - rw_msg->pvnr_resid;
2723
2724 bp->b_resid = bp->b_bcount - moved;
2725 if (bp->b_resid != 0) {
2726 ERROUT(EIO);
2727 }
2728 }
2729
2730 out:
2731 if (park_rw)
2732 puffs_msgmem_release(park_rw);
2733
2734 if (error)
2735 bp->b_error = error;
2736
2737 if (error || dobiodone)
2738 biodone(bp);
2739
2740 return error;
2741 }
2742
2743 int
2744 puffs_vnop_mmap(void *v)
2745 {
2746 struct vop_mmap_args /* {
2747 const struct vnodeop_desc *a_desc;
2748 struct vnode *a_vp;
2749 vm_prot_t a_prot;
2750 kauth_cred_t a_cred;
2751 } */ *ap = v;
2752 PUFFS_MSG_VARS(vn, mmap);
2753 struct vnode *vp = ap->a_vp;
2754 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2755 int error;
2756
2757 if (!PUFFS_USE_PAGECACHE(pmp))
2758 return genfs_eopnotsupp(v);
2759
2760 if (EXISTSOP(pmp, MMAP)) {
2761 PUFFS_MSG_ALLOC(vn, mmap);
2762 mmap_msg->pvnr_prot = ap->a_prot;
2763 puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2764 puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2765 PUFFS_VN_MMAP, VPTOPNC(vp));
2766
2767 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2768 error = checkerr(pmp, error, __func__);
2769 PUFFS_MSG_RELEASE(mmap);
2770 } else {
2771 error = genfs_mmap(v);
2772 }
2773
2774 return error;
2775 }
2776
2777
2778 /*
2779 * The rest don't get a free trip to userspace and back, they
2780 * have to stay within the kernel.
2781 */
2782
2783 /*
2784 * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2785 * well, maybe somehow, somewhere, some day ....
2786 */
2787 int
2788 puffs_vnop_bmap(void *v)
2789 {
2790 struct vop_bmap_args /* {
2791 const struct vnodeop_desc *a_desc;
2792 struct vnode *a_vp;
2793 daddr_t a_bn;
2794 struct vnode **a_vpp;
2795 daddr_t *a_bnp;
2796 int *a_runp;
2797 } */ *ap = v;
2798 struct puffs_mount *pmp;
2799
2800 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2801
2802 if (ap->a_vpp)
2803 *ap->a_vpp = ap->a_vp;
2804 if (ap->a_bnp)
2805 *ap->a_bnp = ap->a_bn;
2806 if (ap->a_runp)
2807 *ap->a_runp
2808 = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
2809
2810 return 0;
2811 }
2812
2813 /*
2814 * Handle getpages faults in puffs. We let genfs_getpages() do most
2815 * of the dirty work, but we come in this route to do accounting tasks.
2816 * If the user server has specified functions for cache notifications
2817 * about reads and/or writes, we record which type of operation we got,
2818 * for which page range, and proceed to issue a FAF notification to the
2819 * server about it.
2820 */
2821 int
2822 puffs_vnop_getpages(void *v)
2823 {
2824 struct vop_getpages_args /* {
2825 const struct vnodeop_desc *a_desc;
2826 struct vnode *a_vp;
2827 voff_t a_offset;
2828 struct vm_page **a_m;
2829 int *a_count;
2830 int a_centeridx;
2831 vm_prot_t a_access_type;
2832 int a_advice;
2833 int a_flags;
2834 } */ *ap = v;
2835 struct puffs_mount *pmp;
2836 struct puffs_node *pn;
2837 struct vnode *vp;
2838 struct vm_page **pgs;
2839 struct puffs_cacheinfo *pcinfo = NULL;
2840 struct puffs_cacherun *pcrun;
2841 void *parkmem = NULL;
2842 size_t runsizes;
2843 int i, npages, si, streakon;
2844 int error, locked, write;
2845
2846 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2847 npages = *ap->a_count;
2848 pgs = ap->a_m;
2849 vp = ap->a_vp;
2850 pn = vp->v_data;
2851 locked = (ap->a_flags & PGO_LOCKED) != 0;
2852 write = (ap->a_access_type & VM_PROT_WRITE) != 0;
2853
2854 /* ccg xnaht - gets Wuninitialized wrong */
2855 pcrun = NULL;
2856 runsizes = 0;
2857
2858 /*
2859 * Check that we aren't trying to fault in pages which our file
2860 * server doesn't know about. This happens if we extend a file by
2861 * skipping some pages and later try to fault in pages which
2862 * are between pn_serversize and vp_size. This check optimizes
2863 * away the common case where a file is being extended.
2864 */
2865 if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
2866 struct vattr va;
2867
2868 /* try again later when we can block */
2869 if (locked)
2870 ERROUT(EBUSY);
2871
2872 mutex_exit(vp->v_interlock);
2873 vattr_null(&va);
2874 va.va_size = vp->v_size;
2875 error = dosetattr(vp, &va, FSCRED, 0);
2876 if (error)
2877 ERROUT(error);
2878 mutex_enter(vp->v_interlock);
2879 }
2880
2881 if (write && PUFFS_WCACHEINFO(pmp)) {
2882 #ifdef notnowjohn
2883 /* allocate worst-case memory */
2884 runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
2885 KASSERT(curlwp != uvm.pagedaemon_lwp || locked);
2886 pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
2887 locked ? KM_NOSLEEP : KM_SLEEP);
2888
2889 /*
2890 * can't block if we're locked and can't mess up caching
2891 * information for fs server. so come back later, please
2892 */
2893 if (pcinfo == NULL)
2894 ERROUT(ENOMEM);
2895
2896 parkmem = puffs_park_alloc(locked == 0);
2897 if (parkmem == NULL)
2898 ERROUT(ENOMEM);
2899
2900 pcrun = pcinfo->pcache_runs;
2901 #else
2902 (void)parkmem;
2903 #endif
2904 }
2905
2906 error = genfs_getpages(v);
2907 if (error)
2908 goto out;
2909
2910 if (PUFFS_WCACHEINFO(pmp) == 0)
2911 goto out;
2912
2913 /*
2914 * Let's see whose fault it was and inform the user server of
2915 * possibly read/written pages. Map pages from read faults
2916 * strictly read-only, since otherwise we might miss info on
2917 * when the page is actually write-faulted to.
2918 */
2919 if (!locked)
2920 mutex_enter(vp->v_uobj.vmobjlock);
2921 for (i = 0, si = 0, streakon = 0; i < npages; i++) {
2922 if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
2923 if (streakon && write) {
2924 streakon = 0;
2925 pcrun[si].pcache_runend
2926 = trunc_page(pgs[i]->offset) + PAGE_MASK;
2927 si++;
2928 }
2929 continue;
2930 }
2931 if (streakon == 0 && write) {
2932 streakon = 1;
2933 pcrun[si].pcache_runstart = pgs[i]->offset;
2934 }
2935
2936 if (!write)
2937 pgs[i]->flags |= PG_RDONLY;
2938 }
2939 /* was the last page part of our streak? */
2940 if (streakon) {
2941 pcrun[si].pcache_runend
2942 = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
2943 si++;
2944 }
2945 if (!locked)
2946 mutex_exit(vp->v_uobj.vmobjlock);
2947
2948 KASSERT(si <= (npages / 2) + 1);
2949
2950 #ifdef notnowjohn
2951 /* send results to userspace */
2952 if (write)
2953 puffs_cacheop(pmp, parkmem, pcinfo,
2954 sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
2955 #endif
2956
2957 out:
2958 if (error) {
2959 if (pcinfo != NULL)
2960 kmem_free(pcinfo,
2961 sizeof(struct puffs_cacheinfo) + runsizes);
2962 #ifdef notnowjohn
2963 if (parkmem != NULL)
2964 puffs_park_release(parkmem, 1);
2965 #endif
2966 }
2967
2968 return error;
2969 }
2970
2971 /*
2972 * Extended attribute support.
2973 */
2974
2975 int
2976 puffs_vnop_getextattr(void *v)
2977 {
2978 struct vop_getextattr_args /*
2979 struct vnode *a_vp;
2980 int a_attrnamespace;
2981 const char *a_name;
2982 struct uio *a_uio;
2983 size_t *a_size;
2984 kauth_cred_t a_cred;
2985 }; */ *ap = v;
2986 PUFFS_MSG_VARS(vn, getextattr);
2987 struct vnode *vp = ap->a_vp;
2988 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2989 int attrnamespace = ap->a_attrnamespace;
2990 const char *name = ap->a_name;
2991 struct uio *uio = ap->a_uio;
2992 size_t *sizep = ap->a_size;
2993 size_t tomove, resid;
2994 int error;
2995
2996 if (uio)
2997 resid = uio->uio_resid;
2998 else
2999 resid = 0;
3000
3001 tomove = PUFFS_TOMOVE(resid, pmp);
3002 if (tomove != resid) {
3003 error = E2BIG;
3004 goto out;
3005 }
3006
3007 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove,
3008 &park_getextattr, (void *)&getextattr_msg, 1);
3009
3010 getextattr_msg->pvnr_attrnamespace = attrnamespace;
3011 strlcpy(getextattr_msg->pvnr_attrname, name,
3012 sizeof(getextattr_msg->pvnr_attrname));
3013 puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred);
3014 if (sizep)
3015 getextattr_msg->pvnr_datasize = 1;
3016 getextattr_msg->pvnr_resid = tomove;
3017
3018 puffs_msg_setinfo(park_getextattr,
3019 PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp));
3020 puffs_msg_setdelta(park_getextattr, tomove);
3021 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error);
3022
3023 error = checkerr(pmp, error, __func__);
3024 if (error)
3025 goto out;
3026
3027 resid = getextattr_msg->pvnr_resid;
3028 if (resid > tomove) {
3029 puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG,
3030 "resid grew", VPTOPNC(vp));
3031 error = EPROTO;
3032 goto out;
3033 }
3034
3035 if (sizep)
3036 *sizep = getextattr_msg->pvnr_datasize;
3037 if (uio)
3038 error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio);
3039
3040 out:
3041 PUFFS_MSG_RELEASE(getextattr);
3042 return error;
3043 }
3044
3045 int
3046 puffs_vnop_setextattr(void *v)
3047 {
3048 struct vop_setextattr_args /* {
3049 struct vnode *a_vp;
3050 int a_attrnamespace;
3051 const char *a_name;
3052 struct uio *a_uio;
3053 kauth_cred_t a_cred;
3054 }; */ *ap = v;
3055 PUFFS_MSG_VARS(vn, setextattr);
3056 struct vnode *vp = ap->a_vp;
3057 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3058 int attrnamespace = ap->a_attrnamespace;
3059 const char *name = ap->a_name;
3060 struct uio *uio = ap->a_uio;
3061 size_t tomove, resid;
3062 int error;
3063
3064 if (uio)
3065 resid = uio->uio_resid;
3066 else
3067 resid = 0;
3068
3069 tomove = PUFFS_TOMOVE(resid, pmp);
3070 if (tomove != resid) {
3071 error = E2BIG;
3072 goto out;
3073 }
3074
3075 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove,
3076 &park_setextattr, (void *)&setextattr_msg, 1);
3077
3078 setextattr_msg->pvnr_attrnamespace = attrnamespace;
3079 strlcpy(setextattr_msg->pvnr_attrname, name,
3080 sizeof(setextattr_msg->pvnr_attrname));
3081 puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred);
3082 setextattr_msg->pvnr_resid = tomove;
3083
3084 if (uio) {
3085 error = uiomove(setextattr_msg->pvnr_data, tomove, uio);
3086 if (error)
3087 goto out;
3088 }
3089
3090 puffs_msg_setinfo(park_setextattr,
3091 PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp));
3092 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error);
3093
3094 error = checkerr(pmp, error, __func__);
3095 if (error)
3096 goto out;
3097
3098 if (setextattr_msg->pvnr_resid != 0)
3099 error = EIO;
3100
3101 out:
3102 PUFFS_MSG_RELEASE(setextattr);
3103
3104 return error;
3105 }
3106
3107 int
3108 puffs_vnop_listextattr(void *v)
3109 {
3110 struct vop_listextattr_args /* {
3111 struct vnode *a_vp;
3112 int a_attrnamespace;
3113 struct uio *a_uio;
3114 size_t *a_size;
3115 int a_flag,
3116 kauth_cred_t a_cred;
3117 }; */ *ap = v;
3118 PUFFS_MSG_VARS(vn, listextattr);
3119 struct vnode *vp = ap->a_vp;
3120 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3121 int attrnamespace = ap->a_attrnamespace;
3122 struct uio *uio = ap->a_uio;
3123 size_t *sizep = ap->a_size;
3124 int flag = ap->a_flag;
3125 size_t tomove, resid;
3126 int error;
3127
3128 if (uio)
3129 resid = uio->uio_resid;
3130 else
3131 resid = 0;
3132
3133 tomove = PUFFS_TOMOVE(resid, pmp);
3134 if (tomove != resid) {
3135 error = E2BIG;
3136 goto out;
3137 }
3138
3139 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_listextattr) + tomove,
3140 &park_listextattr, (void *)&listextattr_msg, 1);
3141
3142 listextattr_msg->pvnr_attrnamespace = attrnamespace;
3143 listextattr_msg->pvnr_flag = flag;
3144 puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred);
3145 listextattr_msg->pvnr_resid = tomove;
3146 if (sizep)
3147 listextattr_msg->pvnr_datasize = 1;
3148
3149 puffs_msg_setinfo(park_listextattr,
3150 PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp));
3151 puffs_msg_setdelta(park_listextattr, tomove);
3152 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error);
3153
3154 error = checkerr(pmp, error, __func__);
3155 if (error)
3156 goto out;
3157
3158 resid = listextattr_msg->pvnr_resid;
3159 if (resid > tomove) {
3160 puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG,
3161 "resid grew", VPTOPNC(vp));
3162 error = EPROTO;
3163 goto out;
3164 }
3165
3166 if (sizep)
3167 *sizep = listextattr_msg->pvnr_datasize;
3168 if (uio)
3169 error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio);
3170
3171 out:
3172 PUFFS_MSG_RELEASE(listextattr);
3173 return error;
3174 }
3175
3176 int
3177 puffs_vnop_deleteextattr(void *v)
3178 {
3179 struct vop_deleteextattr_args /* {
3180 struct vnode *a_vp;
3181 int a_attrnamespace;
3182 const char *a_name;
3183 kauth_cred_t a_cred;
3184 }; */ *ap = v;
3185 PUFFS_MSG_VARS(vn, deleteextattr);
3186 struct vnode *vp = ap->a_vp;
3187 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3188 int attrnamespace = ap->a_attrnamespace;
3189 const char *name = ap->a_name;
3190 int error;
3191
3192 PUFFS_MSG_ALLOC(vn, deleteextattr);
3193 deleteextattr_msg->pvnr_attrnamespace = attrnamespace;
3194 strlcpy(deleteextattr_msg->pvnr_attrname, name,
3195 sizeof(deleteextattr_msg->pvnr_attrname));
3196 puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred);
3197
3198 puffs_msg_setinfo(park_deleteextattr,
3199 PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp));
3200 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr,
3201 vp->v_data, NULL, error);
3202
3203 error = checkerr(pmp, error, __func__);
3204
3205 PUFFS_MSG_RELEASE(deleteextattr);
3206 return error;
3207 }
3208
3209 /*
3210 * spec & fifo. These call the miscfs spec and fifo vectors, but issue
3211 * FAF update information for the puffs node first.
3212 */
3213 int
3214 puffs_vnop_spec_read(void *v)
3215 {
3216 struct vop_read_args /* {
3217 const struct vnodeop_desc *a_desc;
3218 struct vnode *a_vp;
3219 struct uio *a_uio;
3220 int a_ioflag;
3221 kauth_cred_t a_cred;
3222 } */ *ap = v;
3223
3224 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3225 return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
3226 }
3227
3228 int
3229 puffs_vnop_spec_write(void *v)
3230 {
3231 struct vop_write_args /* {
3232 const struct vnodeop_desc *a_desc;
3233 struct vnode *a_vp;
3234 struct uio *a_uio;
3235 int a_ioflag;
3236 kauth_cred_t a_cred;
3237 } */ *ap = v;
3238
3239 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3240 return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
3241 }
3242
3243 int
3244 puffs_vnop_fifo_read(void *v)
3245 {
3246 struct vop_read_args /* {
3247 const struct vnodeop_desc *a_desc;
3248 struct vnode *a_vp;
3249 struct uio *a_uio;
3250 int a_ioflag;
3251 kauth_cred_t a_cred;
3252 } */ *ap = v;
3253
3254 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3255 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
3256 }
3257
3258 int
3259 puffs_vnop_fifo_write(void *v)
3260 {
3261 struct vop_write_args /* {
3262 const struct vnodeop_desc *a_desc;
3263 struct vnode *a_vp;
3264 struct uio *a_uio;
3265 int a_ioflag;
3266 kauth_cred_t a_cred;
3267 } */ *ap = v;
3268
3269 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3270 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
3271 }
3272