puffs_vnops.c revision 1.176 1 /* $NetBSD: puffs_vnops.c,v 1.176 2012/11/05 17:27:38 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 1.176 2012/11/05 17:27:38 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_fsync_desc, puffs_vnop_fsync }, /* REAL fsync */
117 { &vop_seek_desc, puffs_vnop_checkop }, /* seek */
118 { &vop_remove_desc, puffs_vnop_checkop }, /* remove */
119 { &vop_link_desc, puffs_vnop_checkop }, /* link */
120 { &vop_rename_desc, puffs_vnop_checkop }, /* rename */
121 { &vop_mkdir_desc, puffs_vnop_checkop }, /* mkdir */
122 { &vop_rmdir_desc, puffs_vnop_checkop }, /* rmdir */
123 { &vop_symlink_desc, puffs_vnop_checkop }, /* symlink */
124 { &vop_readdir_desc, puffs_vnop_checkop }, /* readdir */
125 { &vop_readlink_desc, puffs_vnop_checkop }, /* readlink */
126 { &vop_getpages_desc, puffs_vnop_checkop }, /* getpages */
127 { &vop_putpages_desc, genfs_putpages }, /* REAL putpages */
128 { &vop_pathconf_desc, puffs_vnop_checkop }, /* pathconf */
129 { &vop_advlock_desc, puffs_vnop_advlock }, /* advlock */
130 { &vop_strategy_desc, puffs_vnop_strategy }, /* REAL strategy */
131 { &vop_revoke_desc, genfs_revoke }, /* REAL revoke */
132 { &vop_abortop_desc, puffs_vnop_abortop }, /* REAL abortop */
133 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
134 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
135 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
136 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
137 { &vop_bmap_desc, puffs_vnop_bmap }, /* REAL bmap */
138 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
139 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
140 { &vop_bwrite_desc, genfs_nullop }, /* REAL bwrite */
141 { &vop_mmap_desc, puffs_vnop_mmap }, /* REAL mmap */
142 { &vop_poll_desc, puffs_vnop_poll }, /* REAL poll */
143 { &vop_getextattr_desc, puffs_vnop_getextattr }, /* getextattr */
144 { &vop_setextattr_desc, puffs_vnop_setextattr }, /* setextattr */
145 { &vop_listextattr_desc, puffs_vnop_listextattr }, /* listextattr */
146 { &vop_deleteextattr_desc, puffs_vnop_deleteextattr },/* deleteextattr */
147 #if 0
148 { &vop_openextattr_desc, puffs_vnop_checkop }, /* openextattr */
149 { &vop_closeextattr_desc, puffs_vnop_checkop }, /* closeextattr */
150 #endif
151 { &vop_kqfilter_desc, genfs_eopnotsupp }, /* kqfilter XXX */
152 { NULL, NULL }
153 };
154 const struct vnodeopv_desc puffs_vnodeop_opv_desc =
155 { &puffs_vnodeop_p, puffs_vnodeop_entries };
156
157
158 int (**puffs_specop_p)(void *);
159 const struct vnodeopv_entry_desc puffs_specop_entries[] = {
160 { &vop_default_desc, vn_default_error },
161 { &vop_lookup_desc, spec_lookup }, /* lookup, ENOTDIR */
162 { &vop_create_desc, spec_create }, /* genfs_badop */
163 { &vop_mknod_desc, spec_mknod }, /* genfs_badop */
164 { &vop_open_desc, spec_open }, /* spec_open */
165 { &vop_close_desc, spec_close }, /* spec_close */
166 { &vop_access_desc, puffs_vnop_checkop }, /* access */
167 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
168 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
169 { &vop_read_desc, puffs_vnop_spec_read }, /* update, read */
170 { &vop_write_desc, puffs_vnop_spec_write }, /* update, write */
171 { &vop_ioctl_desc, spec_ioctl }, /* spec_ioctl */
172 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
173 { &vop_poll_desc, spec_poll }, /* spec_poll */
174 { &vop_kqfilter_desc, spec_kqfilter }, /* spec_kqfilter */
175 { &vop_revoke_desc, spec_revoke }, /* genfs_revoke */
176 { &vop_mmap_desc, spec_mmap }, /* spec_mmap */
177 { &vop_fsync_desc, spec_fsync }, /* vflushbuf */
178 { &vop_seek_desc, spec_seek }, /* genfs_nullop */
179 { &vop_remove_desc, spec_remove }, /* genfs_badop */
180 { &vop_link_desc, spec_link }, /* genfs_badop */
181 { &vop_rename_desc, spec_rename }, /* genfs_badop */
182 { &vop_mkdir_desc, spec_mkdir }, /* genfs_badop */
183 { &vop_rmdir_desc, spec_rmdir }, /* genfs_badop */
184 { &vop_symlink_desc, spec_symlink }, /* genfs_badop */
185 { &vop_readdir_desc, spec_readdir }, /* genfs_badop */
186 { &vop_readlink_desc, spec_readlink }, /* genfs_badop */
187 { &vop_abortop_desc, spec_abortop }, /* genfs_badop */
188 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
189 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
190 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
191 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
192 { &vop_bmap_desc, spec_bmap }, /* dummy */
193 { &vop_strategy_desc, spec_strategy }, /* dev strategy */
194 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
195 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
196 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
197 { &vop_advlock_desc, spec_advlock }, /* lf_advlock */
198 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
199 { &vop_getpages_desc, spec_getpages }, /* genfs_getpages */
200 { &vop_putpages_desc, spec_putpages }, /* genfs_putpages */
201 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
202 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
203 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
204 { &vop_deleteextattr_desc, puffs_vnop_checkop },/* deleteextattr */
205 #if 0
206 { &vop_openextattr_desc, _openextattr }, /* openextattr */
207 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
208 #endif
209 { NULL, NULL }
210 };
211 const struct vnodeopv_desc puffs_specop_opv_desc =
212 { &puffs_specop_p, puffs_specop_entries };
213
214
215 int (**puffs_fifoop_p)(void *);
216 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = {
217 { &vop_default_desc, vn_default_error },
218 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup, ENOTDIR */
219 { &vop_create_desc, vn_fifo_bypass }, /* genfs_badop */
220 { &vop_mknod_desc, vn_fifo_bypass }, /* genfs_badop */
221 { &vop_open_desc, vn_fifo_bypass }, /* open */
222 { &vop_close_desc, vn_fifo_bypass }, /* close */
223 { &vop_access_desc, puffs_vnop_checkop }, /* access */
224 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */
225 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */
226 { &vop_read_desc, puffs_vnop_fifo_read }, /* read, update */
227 { &vop_write_desc, puffs_vnop_fifo_write }, /* write, update */
228 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
229 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */
230 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
231 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
232 { &vop_revoke_desc, vn_fifo_bypass }, /* genfs_revoke */
233 { &vop_mmap_desc, vn_fifo_bypass }, /* genfs_badop */
234 { &vop_fsync_desc, vn_fifo_bypass }, /* genfs_nullop*/
235 { &vop_seek_desc, vn_fifo_bypass }, /* genfs_badop */
236 { &vop_remove_desc, vn_fifo_bypass }, /* genfs_badop */
237 { &vop_link_desc, vn_fifo_bypass }, /* genfs_badop */
238 { &vop_rename_desc, vn_fifo_bypass }, /* genfs_badop */
239 { &vop_mkdir_desc, vn_fifo_bypass }, /* genfs_badop */
240 { &vop_rmdir_desc, vn_fifo_bypass }, /* genfs_badop */
241 { &vop_symlink_desc, vn_fifo_bypass }, /* genfs_badop */
242 { &vop_readdir_desc, vn_fifo_bypass }, /* genfs_badop */
243 { &vop_readlink_desc, vn_fifo_bypass }, /* genfs_badop */
244 { &vop_abortop_desc, vn_fifo_bypass }, /* genfs_badop */
245 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */
246 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */
247 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */
248 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */
249 { &vop_bmap_desc, vn_fifo_bypass }, /* dummy */
250 { &vop_strategy_desc, vn_fifo_bypass }, /* genfs_badop */
251 { &vop_print_desc, puffs_vnop_print }, /* REAL print */
252 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */
253 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
254 { &vop_advlock_desc, vn_fifo_bypass }, /* genfs_einval */
255 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
256 { &vop_putpages_desc, vn_fifo_bypass }, /* genfs_null_putpages*/
257 #if 0
258 { &vop_openextattr_desc, _openextattr }, /* openextattr */
259 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */
260 #endif
261 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */
262 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */
263 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */
264 { &vop_deleteextattr_desc, puffs_vnop_checkop }, /* deleteextattr */
265 { NULL, NULL }
266 };
267 const struct vnodeopv_desc puffs_fifoop_opv_desc =
268 { &puffs_fifoop_p, puffs_fifoop_entries };
269
270
271 /* "real" vnode operations */
272 int (**puffs_msgop_p)(void *);
273 const struct vnodeopv_entry_desc puffs_msgop_entries[] = {
274 { &vop_default_desc, vn_default_error },
275 { &vop_create_desc, puffs_vnop_create }, /* create */
276 { &vop_mknod_desc, puffs_vnop_mknod }, /* mknod */
277 { &vop_open_desc, puffs_vnop_open }, /* open */
278 { &vop_close_desc, puffs_vnop_close }, /* close */
279 { &vop_access_desc, puffs_vnop_access }, /* access */
280 { &vop_getattr_desc, puffs_vnop_getattr }, /* getattr */
281 { &vop_setattr_desc, puffs_vnop_setattr }, /* setattr */
282 { &vop_read_desc, puffs_vnop_read }, /* read */
283 { &vop_write_desc, puffs_vnop_write }, /* write */
284 { &vop_seek_desc, puffs_vnop_seek }, /* seek */
285 { &vop_remove_desc, puffs_vnop_remove }, /* remove */
286 { &vop_link_desc, puffs_vnop_link }, /* link */
287 { &vop_rename_desc, puffs_vnop_rename }, /* rename */
288 { &vop_mkdir_desc, puffs_vnop_mkdir }, /* mkdir */
289 { &vop_rmdir_desc, puffs_vnop_rmdir }, /* rmdir */
290 { &vop_symlink_desc, puffs_vnop_symlink }, /* symlink */
291 { &vop_readdir_desc, puffs_vnop_readdir }, /* readdir */
292 { &vop_readlink_desc, puffs_vnop_readlink }, /* readlink */
293 { &vop_print_desc, puffs_vnop_print }, /* print */
294 { &vop_islocked_desc, puffs_vnop_islocked }, /* islocked */
295 { &vop_pathconf_desc, puffs_vnop_pathconf }, /* pathconf */
296 { &vop_getpages_desc, puffs_vnop_getpages }, /* getpages */
297 { NULL, NULL }
298 };
299 const struct vnodeopv_desc puffs_msgop_opv_desc =
300 { &puffs_msgop_p, puffs_msgop_entries };
301
302 /*
303 * for dosetattr / update_va
304 */
305 #define SETATTR_CHSIZE 0x01
306 #define SETATTR_ASYNC 0x02
307
308 #define ERROUT(err) \
309 do { \
310 error = err; \
311 goto out; \
312 } while (/*CONSTCOND*/0)
313
314 /*
315 * This is a generic vnode operation handler. It checks if the necessary
316 * operations for the called vnode operation are implemented by userspace
317 * and either returns a dummy return value or proceeds to call the real
318 * vnode operation from puffs_msgop_v.
319 *
320 * XXX: this should described elsewhere and autogenerated, the complexity
321 * of the vnode operations vectors and their interrelationships is also
322 * getting a bit out of hand. Another problem is that we need this same
323 * information in the fs server code, so keeping the two in sync manually
324 * is not a viable (long term) plan.
325 */
326
327 /* not supported, handle locking protocol */
328 #define CHECKOP_NOTSUPP(op) \
329 case VOP_##op##_DESCOFFSET: \
330 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
331 return genfs_eopnotsupp(v); \
332 break
333
334 /* always succeed, no locking */
335 #define CHECKOP_SUCCESS(op) \
336 case VOP_##op##_DESCOFFSET: \
337 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \
338 return 0; \
339 break
340
341 int
342 puffs_vnop_checkop(void *v)
343 {
344 struct vop_generic_args /* {
345 struct vnodeop_desc *a_desc;
346 spooky mystery contents;
347 } */ *ap = v;
348 struct vnodeop_desc *desc = ap->a_desc;
349 struct puffs_mount *pmp;
350 struct vnode *vp;
351 int offset, rv;
352
353 offset = ap->a_desc->vdesc_vp_offsets[0];
354 #ifdef DIAGNOSTIC
355 if (offset == VDESC_NO_OFFSET)
356 panic("puffs_checkop: no vnode, why did you call me?");
357 #endif
358 vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap);
359 pmp = MPTOPUFFSMP(vp->v_mount);
360
361 DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n",
362 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp));
363
364 if (!ALLOPS(pmp)) {
365 switch (desc->vdesc_offset) {
366 CHECKOP_NOTSUPP(CREATE);
367 CHECKOP_NOTSUPP(MKNOD);
368 CHECKOP_NOTSUPP(GETATTR);
369 CHECKOP_NOTSUPP(SETATTR);
370 CHECKOP_NOTSUPP(READ);
371 CHECKOP_NOTSUPP(WRITE);
372 CHECKOP_NOTSUPP(FCNTL);
373 CHECKOP_NOTSUPP(IOCTL);
374 CHECKOP_NOTSUPP(REMOVE);
375 CHECKOP_NOTSUPP(LINK);
376 CHECKOP_NOTSUPP(RENAME);
377 CHECKOP_NOTSUPP(MKDIR);
378 CHECKOP_NOTSUPP(RMDIR);
379 CHECKOP_NOTSUPP(SYMLINK);
380 CHECKOP_NOTSUPP(READDIR);
381 CHECKOP_NOTSUPP(READLINK);
382 CHECKOP_NOTSUPP(PRINT);
383 CHECKOP_NOTSUPP(PATHCONF);
384 CHECKOP_NOTSUPP(GETEXTATTR);
385 CHECKOP_NOTSUPP(SETEXTATTR);
386 CHECKOP_NOTSUPP(LISTEXTATTR);
387 CHECKOP_NOTSUPP(DELETEEXTATTR);
388
389 CHECKOP_SUCCESS(ACCESS);
390 CHECKOP_SUCCESS(CLOSE);
391 CHECKOP_SUCCESS(SEEK);
392
393 case VOP_GETPAGES_DESCOFFSET:
394 if (!EXISTSOP(pmp, READ))
395 return genfs_eopnotsupp(v);
396 break;
397
398 default:
399 panic("puffs_checkop: unhandled vnop %d",
400 desc->vdesc_offset);
401 }
402 }
403
404 rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v);
405
406 DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n",
407 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv));
408
409 return rv;
410 }
411
412 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
413 struct componentname *);
414 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t,
415 struct componentname *);
416 static void callinactive(struct puffs_mount *, puffs_cookie_t, int);
417 static void callreclaim(struct puffs_mount *, puffs_cookie_t, int);
418 static int flushvncache(struct vnode *, off_t, off_t, bool);
419 static void update_va(struct vnode *, struct vattr *, struct vattr *,
420 struct timespec *, struct timespec *, int);
421 static void update_parent(struct vnode *, struct vnode *);
422
423
424 #define PUFFS_ABORT_LOOKUP 1
425 #define PUFFS_ABORT_CREATE 2
426 #define PUFFS_ABORT_MKNOD 3
427 #define PUFFS_ABORT_MKDIR 4
428 #define PUFFS_ABORT_SYMLINK 5
429
430 /*
431 * Press the pani^Wabort button! Kernel resource allocation failed.
432 */
433 static void
434 puffs_abortbutton(struct puffs_mount *pmp, int what,
435 puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp)
436 {
437
438 switch (what) {
439 case PUFFS_ABORT_CREATE:
440 case PUFFS_ABORT_MKNOD:
441 case PUFFS_ABORT_SYMLINK:
442 callremove(pmp, dck, ck, cnp);
443 break;
444 case PUFFS_ABORT_MKDIR:
445 callrmdir(pmp, dck, ck, cnp);
446 break;
447 }
448
449 callinactive(pmp, ck, 0);
450 callreclaim(pmp, ck, 1);
451 }
452
453 /*
454 * Begin vnode operations.
455 *
456 * A word from the keymaster about locks: generally we don't want
457 * to use the vnode locks at all: it creates an ugly dependency between
458 * the userlandia file server and the kernel. But we'll play along with
459 * the kernel vnode locks for now. However, even currently we attempt
460 * to release locks as early as possible. This is possible for some
461 * operations which a) don't need a locked vnode after the userspace op
462 * and b) return with the vnode unlocked. Theoretically we could
463 * unlock-do op-lock for others and order the graph in userspace, but I
464 * don't want to think of the consequences for the time being.
465 */
466
467 #define TTL_TO_TIMEOUT(ts) \
468 (hardclock_ticks + (ts->tv_sec * hz) + (ts->tv_nsec * hz / 1000000000))
469 #define TTL_VALID(ts) \
470 ((ts != NULL) && !((ts->tv_sec == 0) && (ts->tv_nsec == 0)))
471 #define TIMED_OUT(expire) \
472 ((int)((unsigned int)hardclock_ticks - (unsigned int)expire) > 0)
473 int
474 puffs_vnop_lookup(void *v)
475 {
476 struct vop_lookup_args /* {
477 const struct vnodeop_desc *a_desc;
478 struct vnode *a_dvp;
479 struct vnode **a_vpp;
480 struct componentname *a_cnp;
481 } */ *ap = v;
482 PUFFS_MSG_VARS(vn, lookup);
483 struct puffs_mount *pmp;
484 struct componentname *cnp;
485 struct vnode *vp, *dvp, *cvp;
486 struct puffs_node *dpn, *cpn;
487 int isdot;
488 int error;
489
490 pmp = MPTOPUFFSMP(ap->a_dvp->v_mount);
491 cnp = ap->a_cnp;
492 dvp = ap->a_dvp;
493 cvp = NULL;
494 cpn = NULL;
495 *ap->a_vpp = NULL;
496
497 /* r/o fs? we check create later to handle EEXIST */
498 if ((cnp->cn_flags & ISLASTCN)
499 && (dvp->v_mount->mnt_flag & MNT_RDONLY)
500 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
501 return EROFS;
502
503 isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.';
504
505 DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n",
506 cnp->cn_nameptr, dvp, cnp->cn_nameiop));
507
508 /*
509 * If dotdot cache is enabled, unlock parent, lock ..
510 * (grand-parent) and relock parent.
511 */
512 if (PUFFS_USE_DOTDOTCACHE(pmp) && (cnp->cn_flags & ISDOTDOT)) {
513 VOP_UNLOCK(dvp);
514
515 vp = VPTOPP(ap->a_dvp)->pn_parent;
516 vref(vp);
517
518 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
519 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
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 locked
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 mutex_enter(&cpn->pn_sizemtx);
587
588 PUFFS_MSG_ALLOC(vn, lookup);
589 puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred,
590 cnp, PUFFS_USE_FULLPNBUF(pmp));
591
592 if (cnp->cn_flags & ISDOTDOT)
593 VOP_UNLOCK(dvp);
594
595 puffs_msg_setinfo(park_lookup, PUFFSOP_VN,
596 PUFFS_VN_LOOKUP, VPTOPNC(dvp));
597 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error);
598 DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error));
599
600 /*
601 * In case of error, there is no new vnode to play with, so be
602 * happy with the NULL value given to vpp in the beginning.
603 * Also, check if this really was an error or the target was not
604 * present. Either treat it as a non-error for CREATE/RENAME or
605 * enter the component into the negative name cache (if desired).
606 */
607 if (error) {
608 error = checkerr(pmp, error, __func__);
609 if (error == ENOENT) {
610 /* don't allow to create files on r/o fs */
611 if ((dvp->v_mount->mnt_flag & MNT_RDONLY)
612 && cnp->cn_nameiop == CREATE) {
613 error = EROFS;
614
615 /* adjust values if we are creating */
616 } else if ((cnp->cn_flags & ISLASTCN)
617 && (cnp->cn_nameiop == CREATE
618 || cnp->cn_nameiop == RENAME)) {
619 error = EJUSTRETURN;
620
621 /* save negative cache entry */
622 } else {
623 if (PUFFS_USE_NAMECACHE(pmp) &&
624 !PUFFS_USE_FS_TTL(pmp))
625 cache_enter(dvp, NULL, cnp->cn_nameptr,
626 cnp->cn_namelen, cnp->cn_flags);
627 }
628 }
629 goto out;
630 }
631
632 /*
633 * Check that we don't get our parent node back, that would cause
634 * a pretty obvious deadlock.
635 */
636 dpn = dvp->v_data;
637 if (lookup_msg->pvnr_newnode == dpn->pn_cookie) {
638 puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL,
639 "lookup produced parent cookie", lookup_msg->pvnr_newnode);
640 error = EPROTO;
641 goto out;
642 }
643
644 /*
645 * Check if we looked up the cached vnode
646 */
647 vp = NULL;
648 if (cvp && (VPTOPP(cvp)->pn_cookie == lookup_msg->pvnr_newnode)) {
649 int grace;
650
651 /*
652 * Bump grace time of this node so that it does not get
653 * reclaimed too fast. We try to increase a bit more the
654 * lifetime of busiest * nodes - with some limits.
655 */
656 grace = 10 * puffs_sopreq_expire_timeout;
657 cpn->pn_cn_grace = hardclock_ticks + grace;
658 vp = cvp;
659 }
660
661 /*
662 * No cached vnode available, or the cached vnode does not
663 * match the userland cookie anymore: is the node known?
664 */
665 if (vp == NULL) {
666 error = puffs_cookie2vnode(pmp, lookup_msg->pvnr_newnode,
667 1, 1, &vp);
668 }
669
670 if (error == PUFFS_NOSUCHCOOKIE) {
671 error = puffs_getvnode(dvp->v_mount,
672 lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype,
673 lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp);
674 if (error) {
675 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP,
676 VPTOPNC(dvp), lookup_msg->pvnr_newnode,
677 ap->a_cnp);
678 goto out;
679 }
680
681 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
682 } else if (error) {
683 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP, VPTOPNC(dvp),
684 lookup_msg->pvnr_newnode, ap->a_cnp);
685 goto out;
686 }
687
688 /*
689 * Update cache and TTL
690 */
691 if (PUFFS_USE_FS_TTL(pmp)) {
692 struct timespec *va_ttl = &lookup_msg->pvnr_va_ttl;
693 struct timespec *cn_ttl = &lookup_msg->pvnr_cn_ttl;
694 update_va(vp, NULL, &lookup_msg->pvnr_va,
695 va_ttl, cn_ttl, SETATTR_CHSIZE);
696 }
697
698 KASSERT(lookup_msg->pvnr_newnode == VPTOPP(vp)->pn_cookie);
699 *ap->a_vpp = vp;
700
701 if (PUFFS_USE_NAMECACHE(pmp))
702 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
703 cnp->cn_flags);
704
705 /* XXX */
706 if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0)
707 cnp->cn_flags &= ~REQUIREDIR;
708 if (lookup_msg->pvnr_cn.pkcn_consume)
709 cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume,
710 strlen(cnp->cn_nameptr) - cnp->cn_namelen);
711
712 VPTOPP(vp)->pn_nlookup++;
713
714 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
715 (VPTOPP(vp)->pn_parent != dvp))
716 update_parent(vp, dvp);
717
718 out:
719 if (cvp != NULL) {
720 mutex_exit(&cpn->pn_sizemtx);
721
722 if (error || (cvp != vp))
723 vput(cvp);
724 }
725
726 if (cnp->cn_flags & ISDOTDOT)
727 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
728
729 DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp));
730 PUFFS_MSG_RELEASE(lookup);
731 return error;
732 }
733
734 #define REFPN_AND_UNLOCKVP(a, b) \
735 do { \
736 mutex_enter(&b->pn_mtx); \
737 puffs_referencenode(b); \
738 mutex_exit(&b->pn_mtx); \
739 VOP_UNLOCK(a); \
740 } while (/*CONSTCOND*/0)
741
742 #define REFPN(b) \
743 do { \
744 mutex_enter(&b->pn_mtx); \
745 puffs_referencenode(b); \
746 mutex_exit(&b->pn_mtx); \
747 } while (/*CONSTCOND*/0)
748
749 #define RELEPN_AND_VP(a, b) \
750 do { \
751 puffs_releasenode(b); \
752 vrele(a); \
753 } while (/*CONSTCOND*/0)
754
755 int
756 puffs_vnop_create(void *v)
757 {
758 struct vop_create_args /* {
759 const struct vnodeop_desc *a_desc;
760 struct vnode *a_dvp;
761 struct vnode **a_vpp;
762 struct componentname *a_cnp;
763 struct vattr *a_vap;
764 } */ *ap = v;
765 PUFFS_MSG_VARS(vn, create);
766 struct vnode *dvp = ap->a_dvp;
767 struct puffs_node *dpn = VPTOPP(dvp);
768 struct componentname *cnp = ap->a_cnp;
769 struct mount *mp = dvp->v_mount;
770 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
771 int error;
772
773 DPRINTF(("puffs_create: dvp %p, cnp: %s\n",
774 dvp, ap->a_cnp->cn_nameptr));
775
776 PUFFS_MSG_ALLOC(vn, create);
777 puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred,
778 cnp, PUFFS_USE_FULLPNBUF(pmp));
779 create_msg->pvnr_va = *ap->a_vap;
780 puffs_msg_setinfo(park_create, PUFFSOP_VN,
781 PUFFS_VN_CREATE, VPTOPNC(dvp));
782 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error);
783
784 error = checkerr(pmp, error, __func__);
785 if (error)
786 goto out;
787
788 error = puffs_newnode(mp, dvp, ap->a_vpp,
789 create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0);
790 if (error) {
791 puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie,
792 create_msg->pvnr_newnode, cnp);
793 goto out;
794 }
795
796 if (PUFFS_USE_FS_TTL(pmp)) {
797 struct timespec *va_ttl = &create_msg->pvnr_va_ttl;
798 struct timespec *cn_ttl = &create_msg->pvnr_cn_ttl;
799 struct vattr *rvap = &create_msg->pvnr_va;
800
801 update_va(*ap->a_vpp, NULL, rvap,
802 va_ttl, cn_ttl, SETATTR_CHSIZE);
803 }
804
805 VPTOPP(*ap->a_vpp)->pn_nlookup++;
806
807 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
808 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
809 update_parent(*ap->a_vpp, dvp);
810
811 out:
812 vput(dvp);
813
814 DPRINTF(("puffs_create: return %d\n", error));
815 PUFFS_MSG_RELEASE(create);
816 return error;
817 }
818
819 int
820 puffs_vnop_mknod(void *v)
821 {
822 struct vop_mknod_args /* {
823 const struct vnodeop_desc *a_desc;
824 struct vnode *a_dvp;
825 struct vnode **a_vpp;
826 struct componentname *a_cnp;
827 struct vattr *a_vap;
828 } */ *ap = v;
829 PUFFS_MSG_VARS(vn, mknod);
830 struct vnode *dvp = ap->a_dvp;
831 struct puffs_node *dpn = VPTOPP(dvp);
832 struct componentname *cnp = ap->a_cnp;
833 struct mount *mp = dvp->v_mount;
834 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
835 int error;
836
837 PUFFS_MSG_ALLOC(vn, mknod);
838 puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred,
839 cnp, PUFFS_USE_FULLPNBUF(pmp));
840 mknod_msg->pvnr_va = *ap->a_vap;
841 puffs_msg_setinfo(park_mknod, PUFFSOP_VN,
842 PUFFS_VN_MKNOD, VPTOPNC(dvp));
843
844 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error);
845
846 error = checkerr(pmp, error, __func__);
847 if (error)
848 goto out;
849
850 error = puffs_newnode(mp, dvp, ap->a_vpp,
851 mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type,
852 ap->a_vap->va_rdev);
853 if (error) {
854 puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie,
855 mknod_msg->pvnr_newnode, cnp);
856 goto out;
857 }
858
859 if (PUFFS_USE_FS_TTL(pmp)) {
860 struct timespec *va_ttl = &mknod_msg->pvnr_va_ttl;
861 struct timespec *cn_ttl = &mknod_msg->pvnr_cn_ttl;
862 struct vattr *rvap = &mknod_msg->pvnr_va;
863
864 update_va(*ap->a_vpp, NULL, rvap,
865 va_ttl, cn_ttl, SETATTR_CHSIZE);
866 }
867
868 VPTOPP(*ap->a_vpp)->pn_nlookup++;
869
870 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
871 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
872 update_parent(*ap->a_vpp, dvp);
873
874 out:
875 vput(dvp);
876 PUFFS_MSG_RELEASE(mknod);
877 return error;
878 }
879
880 int
881 puffs_vnop_open(void *v)
882 {
883 struct vop_open_args /* {
884 const struct vnodeop_desc *a_desc;
885 struct vnode *a_vp;
886 int a_mode;
887 kauth_cred_t a_cred;
888 } */ *ap = v;
889 PUFFS_MSG_VARS(vn, open);
890 struct vnode *vp = ap->a_vp;
891 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
892 int mode = ap->a_mode;
893 int error;
894
895 DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode));
896
897 if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE))
898 ERROUT(EROFS);
899
900 if (!EXISTSOP(pmp, OPEN))
901 ERROUT(0);
902
903 PUFFS_MSG_ALLOC(vn, open);
904 open_msg->pvnr_mode = mode;
905 puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred);
906 puffs_msg_setinfo(park_open, PUFFSOP_VN,
907 PUFFS_VN_OPEN, VPTOPNC(vp));
908
909 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error);
910 error = checkerr(pmp, error, __func__);
911
912 out:
913 DPRINTF(("puffs_open: returning %d\n", error));
914 PUFFS_MSG_RELEASE(open);
915 return error;
916 }
917
918 int
919 puffs_vnop_close(void *v)
920 {
921 struct vop_close_args /* {
922 const struct vnodeop_desc *a_desc;
923 struct vnode *a_vp;
924 int a_fflag;
925 kauth_cred_t a_cred;
926 } */ *ap = v;
927 PUFFS_MSG_VARS(vn, close);
928 struct vnode *vp = ap->a_vp;
929 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
930
931 PUFFS_MSG_ALLOC(vn, close);
932 puffs_msg_setfaf(park_close);
933 close_msg->pvnr_fflag = ap->a_fflag;
934 puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred);
935 puffs_msg_setinfo(park_close, PUFFSOP_VN,
936 PUFFS_VN_CLOSE, VPTOPNC(vp));
937
938 puffs_msg_enqueue(pmp, park_close);
939 PUFFS_MSG_RELEASE(close);
940 return 0;
941 }
942
943 int
944 puffs_vnop_access(void *v)
945 {
946 struct vop_access_args /* {
947 const struct vnodeop_desc *a_desc;
948 struct vnode *a_vp;
949 int a_mode;
950 kauth_cred_t a_cred;
951 } */ *ap = v;
952 PUFFS_MSG_VARS(vn, access);
953 struct vnode *vp = ap->a_vp;
954 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
955 int mode = ap->a_mode;
956 int error;
957
958 if (mode & VWRITE) {
959 switch (vp->v_type) {
960 case VDIR:
961 case VLNK:
962 case VREG:
963 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
964 || !EXISTSOP(pmp, WRITE))
965 return EROFS;
966 break;
967 default:
968 break;
969 }
970 }
971
972 if (!EXISTSOP(pmp, ACCESS))
973 return 0;
974
975 PUFFS_MSG_ALLOC(vn, access);
976 access_msg->pvnr_mode = ap->a_mode;
977 puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred);
978 puffs_msg_setinfo(park_access, PUFFSOP_VN,
979 PUFFS_VN_ACCESS, VPTOPNC(vp));
980
981 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error);
982 error = checkerr(pmp, error, __func__);
983 PUFFS_MSG_RELEASE(access);
984
985 return error;
986 }
987
988 static void
989 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap,
990 struct timespec *va_ttl, struct timespec *cn_ttl, int flags)
991 {
992 struct puffs_node *pn = VPTOPP(vp);
993
994 if (TTL_VALID(cn_ttl)) {
995 pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl);
996 pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace);
997 }
998
999 /*
1000 * Don't listen to the file server regarding special device
1001 * size info, the file server doesn't know anything about them.
1002 */
1003 if (vp->v_type == VBLK || vp->v_type == VCHR)
1004 rvap->va_size = vp->v_size;
1005
1006 /* Ditto for blocksize (ufs comment: this doesn't belong here) */
1007 if (vp->v_type == VBLK)
1008 rvap->va_blocksize = BLKDEV_IOSIZE;
1009 else if (vp->v_type == VCHR)
1010 rvap->va_blocksize = MAXBSIZE;
1011
1012 if (vap != NULL) {
1013 (void) memcpy(vap, rvap, sizeof(struct vattr));
1014 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1015
1016 if (pn->pn_stat & PNODE_METACACHE_ATIME)
1017 vap->va_atime = pn->pn_mc_atime;
1018 if (pn->pn_stat & PNODE_METACACHE_CTIME)
1019 vap->va_ctime = pn->pn_mc_ctime;
1020 if (pn->pn_stat & PNODE_METACACHE_MTIME)
1021 vap->va_mtime = pn->pn_mc_mtime;
1022 if (pn->pn_stat & PNODE_METACACHE_SIZE)
1023 vap->va_size = pn->pn_mc_size;
1024 }
1025
1026 if (!(pn->pn_stat & PNODE_METACACHE_SIZE) && (flags & SETATTR_CHSIZE)) {
1027 if (rvap->va_size != VNOVAL
1028 && vp->v_type != VBLK && vp->v_type != VCHR) {
1029 uvm_vnp_setsize(vp, rvap->va_size);
1030 pn->pn_serversize = rvap->va_size;
1031 }
1032 }
1033
1034 if ((va_ttl != NULL) && TTL_VALID(va_ttl)) {
1035 if (pn->pn_va_cache == NULL)
1036 pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK);
1037
1038 (void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap));
1039
1040 pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl);
1041 }
1042 }
1043
1044 static void
1045 update_parent(struct vnode *vp, struct vnode *dvp)
1046 {
1047 struct puffs_node *pn = VPTOPP(vp);
1048
1049 if (pn->pn_parent != NULL) {
1050 KASSERT(pn->pn_parent != dvp);
1051 vrele(pn->pn_parent);
1052 }
1053
1054 vref(dvp);
1055 pn->pn_parent = dvp;
1056 }
1057
1058 int
1059 puffs_vnop_getattr(void *v)
1060 {
1061 struct vop_getattr_args /* {
1062 const struct vnodeop_desc *a_desc;
1063 struct vnode *a_vp;
1064 struct vattr *a_vap;
1065 kauth_cred_t a_cred;
1066 } */ *ap = v;
1067 PUFFS_MSG_VARS(vn, getattr);
1068 struct vnode *vp = ap->a_vp;
1069 struct mount *mp = vp->v_mount;
1070 struct puffs_mount *pmp = MPTOPUFFSMP(mp);
1071 struct vattr *vap, *rvap;
1072 struct puffs_node *pn = VPTOPP(vp);
1073 struct timespec *va_ttl = NULL;
1074 int error = 0;
1075
1076 /*
1077 * A lock is required so that we do not race with
1078 * setattr, write and fsync when changing vp->v_size.
1079 * This is critical, since setting a stall smaler value
1080 * triggers a file truncate in uvm_vnp_setsize(), which
1081 * most of the time means data corruption (a chunk of
1082 * data is replaced by zeroes). This can be removed if
1083 * we decide one day that VOP_GETATTR must operate on
1084 * a locked vnode.
1085 *
1086 * XXX Should be useless now that VOP_GETATTR has been
1087 * fixed to always require a shared lock at least.
1088 */
1089 mutex_enter(&pn->pn_sizemtx);
1090
1091 REFPN(pn);
1092 vap = ap->a_vap;
1093
1094 if (PUFFS_USE_FS_TTL(pmp)) {
1095 if (!TIMED_OUT(pn->pn_va_timeout)) {
1096 update_va(vp, vap, pn->pn_va_cache,
1097 NULL, NULL, SETATTR_CHSIZE);
1098 goto out2;
1099 }
1100 }
1101
1102 PUFFS_MSG_ALLOC(vn, getattr);
1103 vattr_null(&getattr_msg->pvnr_va);
1104 puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred);
1105 puffs_msg_setinfo(park_getattr, PUFFSOP_VN,
1106 PUFFS_VN_GETATTR, VPTOPNC(vp));
1107
1108 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error);
1109 error = checkerr(pmp, error, __func__);
1110 if (error)
1111 goto out;
1112
1113 rvap = &getattr_msg->pvnr_va;
1114
1115 if (PUFFS_USE_FS_TTL(pmp))
1116 va_ttl = &getattr_msg->pvnr_va_ttl;
1117
1118 update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE);
1119
1120 out:
1121 PUFFS_MSG_RELEASE(getattr);
1122
1123 out2:
1124 puffs_releasenode(pn);
1125
1126 mutex_exit(&pn->pn_sizemtx);
1127
1128 return error;
1129 }
1130
1131 static int
1132 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags)
1133 {
1134 PUFFS_MSG_VARS(vn, setattr);
1135 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1136 struct puffs_node *pn = vp->v_data;
1137 int error = 0;
1138
1139 KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx));
1140
1141 if ((vp->v_mount->mnt_flag & MNT_RDONLY) &&
1142 (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL
1143 || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL
1144 || vap->va_mode != (mode_t)VNOVAL))
1145 return EROFS;
1146
1147 if ((vp->v_mount->mnt_flag & MNT_RDONLY)
1148 && vp->v_type == VREG && vap->va_size != VNOVAL)
1149 return EROFS;
1150
1151 /*
1152 * Flush metacache first. If we are called with some explicit
1153 * parameters, treat them as information overriding metacache
1154 * information.
1155 */
1156 if (pn->pn_stat & PNODE_METACACHE_MASK) {
1157 if ((pn->pn_stat & PNODE_METACACHE_ATIME)
1158 && vap->va_atime.tv_sec == VNOVAL)
1159 vap->va_atime = pn->pn_mc_atime;
1160 if ((pn->pn_stat & PNODE_METACACHE_CTIME)
1161 && vap->va_ctime.tv_sec == VNOVAL)
1162 vap->va_ctime = pn->pn_mc_ctime;
1163 if ((pn->pn_stat & PNODE_METACACHE_MTIME)
1164 && vap->va_mtime.tv_sec == VNOVAL)
1165 vap->va_mtime = pn->pn_mc_mtime;
1166 if ((pn->pn_stat & PNODE_METACACHE_SIZE)
1167 && vap->va_size == VNOVAL)
1168 vap->va_size = pn->pn_mc_size;
1169
1170 pn->pn_stat &= ~PNODE_METACACHE_MASK;
1171 }
1172
1173 /*
1174 * Flush attribute cache so that another thread do
1175 * not get a stale value during the operation.
1176 */
1177 if (PUFFS_USE_FS_TTL(pmp))
1178 pn->pn_va_timeout = 0;
1179
1180 PUFFS_MSG_ALLOC(vn, setattr);
1181 (void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr));
1182 puffs_credcvt(&setattr_msg->pvnr_cred, cred);
1183 puffs_msg_setinfo(park_setattr, PUFFSOP_VN,
1184 PUFFS_VN_SETATTR, VPTOPNC(vp));
1185 if (flags & SETATTR_ASYNC)
1186 puffs_msg_setfaf(park_setattr);
1187
1188 puffs_msg_enqueue(pmp, park_setattr);
1189 if ((flags & SETATTR_ASYNC) == 0)
1190 error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL);
1191
1192 if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) {
1193 struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl;
1194 struct vattr *rvap = &setattr_msg->pvnr_va;
1195
1196 update_va(vp, NULL, rvap, va_ttl, NULL, flags);
1197 }
1198
1199 PUFFS_MSG_RELEASE(setattr);
1200 if ((flags & SETATTR_ASYNC) == 0) {
1201 error = checkerr(pmp, error, __func__);
1202 if (error)
1203 return error;
1204 } else {
1205 error = 0;
1206 }
1207
1208 if (vap->va_size != VNOVAL) {
1209 pn->pn_serversize = vap->va_size;
1210 if (flags & SETATTR_CHSIZE)
1211 uvm_vnp_setsize(vp, vap->va_size);
1212 }
1213
1214 return 0;
1215 }
1216
1217 int
1218 puffs_vnop_setattr(void *v)
1219 {
1220 struct vop_getattr_args /* {
1221 const struct vnodeop_desc *a_desc;
1222 struct vnode *a_vp;
1223 struct vattr *a_vap;
1224 kauth_cred_t a_cred;
1225 } */ *ap = v;
1226 struct puffs_node *pn = ap->a_vp->v_data;
1227 int error;
1228
1229 mutex_enter(&pn->pn_sizemtx);
1230 error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE);
1231 mutex_exit(&pn->pn_sizemtx);
1232
1233 return error;
1234 }
1235
1236 static __inline int
1237 doinact(struct puffs_mount *pmp, int iaflag)
1238 {
1239
1240 if (EXISTSOP(pmp, INACTIVE))
1241 if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND)
1242 if (iaflag || ALLOPS(pmp))
1243 return 1;
1244 else
1245 return 0;
1246 else
1247 return 1;
1248 else
1249 return 0;
1250 }
1251
1252 static void
1253 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag)
1254 {
1255 int error;
1256 PUFFS_MSG_VARS(vn, inactive);
1257
1258 if (doinact(pmp, iaflag)) {
1259 PUFFS_MSG_ALLOC(vn, inactive);
1260 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1261 PUFFS_VN_INACTIVE, ck);
1262
1263 PUFFS_MSG_ENQUEUEWAIT(pmp, park_inactive, error);
1264 PUFFS_MSG_RELEASE(inactive);
1265 }
1266 }
1267
1268 /* XXX: callinactive can't setback */
1269 int
1270 puffs_vnop_inactive(void *v)
1271 {
1272 struct vop_inactive_args /* {
1273 const struct vnodeop_desc *a_desc;
1274 struct vnode *a_vp;
1275 } */ *ap = v;
1276 PUFFS_MSG_VARS(vn, inactive);
1277 struct vnode *vp = ap->a_vp;
1278 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
1279 struct puffs_node *pnode;
1280 bool recycle = false;
1281 int error;
1282
1283 pnode = vp->v_data;
1284 mutex_enter(&pnode->pn_sizemtx);
1285
1286 if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) {
1287 flushvncache(vp, 0, 0, false);
1288 PUFFS_MSG_ALLOC(vn, inactive);
1289 puffs_msg_setinfo(park_inactive, PUFFSOP_VN,
1290 PUFFS_VN_INACTIVE, VPTOPNC(vp));
1291
1292 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_inactive, vp->v_data,
1293 NULL, error);
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, error;
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(pmp, park_poll, vp->v_data,
1609 NULL, error);
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 (vp->v_iflag & VI_XLOCK)
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_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 vput(dvp);
1876 PUFFS_MSG_RELEASE(mkdir);
1877 return error;
1878 }
1879
1880 static int
1881 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck,
1882 struct componentname *cnp)
1883 {
1884 PUFFS_MSG_VARS(vn, rmdir);
1885 int error;
1886
1887 PUFFS_MSG_ALLOC(vn, rmdir);
1888 rmdir_msg->pvnr_cookie_targ = ck;
1889 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1890 cnp, PUFFS_USE_FULLPNBUF(pmp));
1891 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck);
1892
1893 PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error);
1894 PUFFS_MSG_RELEASE(rmdir);
1895
1896 return checkerr(pmp, error, __func__);
1897 }
1898
1899 int
1900 puffs_vnop_rmdir(void *v)
1901 {
1902 struct vop_rmdir_args /* {
1903 const struct vnodeop_desc *a_desc;
1904 struct vnode *a_dvp;
1905 struct vnode *a_vp;
1906 struct componentname *a_cnp;
1907 } */ *ap = v;
1908 PUFFS_MSG_VARS(vn, rmdir);
1909 struct vnode *dvp = ap->a_dvp;
1910 struct vnode *vp = ap->a_vp;
1911 struct puffs_node *dpn = VPTOPP(dvp);
1912 struct puffs_node *pn = VPTOPP(vp);
1913 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1914 struct componentname *cnp = ap->a_cnp;
1915 int error;
1916
1917 PUFFS_MSG_ALLOC(vn, rmdir);
1918 rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp);
1919 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred,
1920 cnp, PUFFS_USE_FULLPNBUF(pmp));
1921 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN,
1922 PUFFS_VN_RMDIR, VPTOPNC(dvp));
1923
1924 puffs_msg_enqueue(pmp, park_rmdir);
1925 REFPN_AND_UNLOCKVP(dvp, dpn);
1926 REFPN_AND_UNLOCKVP(vp, pn);
1927 error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn);
1928
1929 PUFFS_MSG_RELEASE(rmdir);
1930
1931 /* XXX: some call cache_purge() *for both vnodes* here, investigate */
1932 RELEPN_AND_VP(dvp, dpn);
1933 RELEPN_AND_VP(vp, pn);
1934
1935 return error;
1936 }
1937
1938 int
1939 puffs_vnop_link(void *v)
1940 {
1941 struct vop_link_args /* {
1942 const struct vnodeop_desc *a_desc;
1943 struct vnode *a_dvp;
1944 struct vnode *a_vp;
1945 struct componentname *a_cnp;
1946 } */ *ap = v;
1947 PUFFS_MSG_VARS(vn, link);
1948 struct vnode *dvp = ap->a_dvp;
1949 struct vnode *vp = ap->a_vp;
1950 struct puffs_node *dpn = VPTOPP(dvp);
1951 struct puffs_node *pn = VPTOPP(vp);
1952 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
1953 struct componentname *cnp = ap->a_cnp;
1954 int error;
1955
1956 PUFFS_MSG_ALLOC(vn, link);
1957 link_msg->pvnr_cookie_targ = VPTOPNC(vp);
1958 puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred,
1959 cnp, PUFFS_USE_FULLPNBUF(pmp));
1960 puffs_msg_setinfo(park_link, PUFFSOP_VN,
1961 PUFFS_VN_LINK, VPTOPNC(dvp));
1962
1963 puffs_msg_enqueue(pmp, park_link);
1964 REFPN_AND_UNLOCKVP(dvp, dpn);
1965 REFPN(pn);
1966 error = puffs_msg_wait2(pmp, park_link, dpn, pn);
1967
1968 PUFFS_MSG_RELEASE(link);
1969
1970 error = checkerr(pmp, error, __func__);
1971
1972 /*
1973 * XXX: stay in touch with the cache. I don't like this, but
1974 * don't have a better solution either. See also puffs_rename().
1975 */
1976 if (error == 0)
1977 puffs_updatenode(pn, PUFFS_UPDATECTIME, 0);
1978
1979 RELEPN_AND_VP(dvp, dpn);
1980 puffs_releasenode(pn);
1981
1982 return error;
1983 }
1984
1985 int
1986 puffs_vnop_symlink(void *v)
1987 {
1988 struct vop_symlink_args /* {
1989 const struct vnodeop_desc *a_desc;
1990 struct vnode *a_dvp;
1991 struct vnode **a_vpp;
1992 struct componentname *a_cnp;
1993 struct vattr *a_vap;
1994 char *a_target;
1995 } */ *ap = v;
1996 PUFFS_MSG_VARS(vn, symlink);
1997 struct vnode *dvp = ap->a_dvp;
1998 struct puffs_node *dpn = VPTOPP(dvp);
1999 struct mount *mp = dvp->v_mount;
2000 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2001 struct componentname *cnp = ap->a_cnp;
2002 int error;
2003
2004 *ap->a_vpp = NULL;
2005
2006 PUFFS_MSG_ALLOC(vn, symlink);
2007 puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred,
2008 cnp, PUFFS_USE_FULLPNBUF(pmp));
2009 symlink_msg->pvnr_va = *ap->a_vap;
2010 (void)strlcpy(symlink_msg->pvnr_link, ap->a_target,
2011 sizeof(symlink_msg->pvnr_link));
2012 puffs_msg_setinfo(park_symlink, PUFFSOP_VN,
2013 PUFFS_VN_SYMLINK, VPTOPNC(dvp));
2014
2015 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error);
2016
2017 error = checkerr(pmp, error, __func__);
2018 if (error)
2019 goto out;
2020
2021 error = puffs_newnode(mp, dvp, ap->a_vpp,
2022 symlink_msg->pvnr_newnode, cnp, VLNK, 0);
2023 if (error) {
2024 puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie,
2025 symlink_msg->pvnr_newnode, cnp);
2026 goto out;
2027 }
2028
2029 if (PUFFS_USE_FS_TTL(pmp)) {
2030 struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl;
2031 struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl;
2032 struct vattr *rvap = &symlink_msg->pvnr_va;
2033
2034 update_va(*ap->a_vpp, NULL, rvap,
2035 va_ttl, cn_ttl, SETATTR_CHSIZE);
2036 }
2037
2038 VPTOPP(*ap->a_vpp)->pn_nlookup++;
2039
2040 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2041 (VPTOPP(*ap->a_vpp)->pn_parent != dvp))
2042 update_parent(*ap->a_vpp, dvp);
2043
2044 out:
2045 vput(dvp);
2046 PUFFS_MSG_RELEASE(symlink);
2047
2048 return error;
2049 }
2050
2051 int
2052 puffs_vnop_readlink(void *v)
2053 {
2054 struct vop_readlink_args /* {
2055 const struct vnodeop_desc *a_desc;
2056 struct vnode *a_vp;
2057 struct uio *a_uio;
2058 kauth_cred_t a_cred;
2059 } */ *ap = v;
2060 PUFFS_MSG_VARS(vn, readlink);
2061 struct vnode *vp = ap->a_vp;
2062 struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2063 size_t linklen;
2064 int error;
2065
2066 PUFFS_MSG_ALLOC(vn, readlink);
2067 puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred);
2068 linklen = sizeof(readlink_msg->pvnr_link);
2069 readlink_msg->pvnr_linklen = linklen;
2070 puffs_msg_setinfo(park_readlink, PUFFSOP_VN,
2071 PUFFS_VN_READLINK, VPTOPNC(vp));
2072
2073 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error);
2074 error = checkerr(pmp, error, __func__);
2075 if (error)
2076 goto out;
2077
2078 /* bad bad user file server */
2079 if (readlink_msg->pvnr_linklen > linklen) {
2080 puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG,
2081 "linklen too big", VPTOPNC(ap->a_vp));
2082 error = EPROTO;
2083 goto out;
2084 }
2085
2086 error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen,
2087 ap->a_uio);
2088 out:
2089 PUFFS_MSG_RELEASE(readlink);
2090 return error;
2091 }
2092
2093 int
2094 puffs_vnop_rename(void *v)
2095 {
2096 struct vop_rename_args /* {
2097 const struct vnodeop_desc *a_desc;
2098 struct vnode *a_fdvp;
2099 struct vnode *a_fvp;
2100 struct componentname *a_fcnp;
2101 struct vnode *a_tdvp;
2102 struct vnode *a_tvp;
2103 struct componentname *a_tcnp;
2104 } */ *ap = v;
2105 PUFFS_MSG_VARS(vn, rename);
2106 struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp;
2107 struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp;
2108 struct puffs_node *fpn = ap->a_fvp->v_data;
2109 struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount);
2110 int error;
2111 bool doabort = true;
2112
2113 if ((fvp->v_mount != tdvp->v_mount) ||
2114 (tvp && (fvp->v_mount != tvp->v_mount))) {
2115 ERROUT(EXDEV);
2116 }
2117
2118 PUFFS_MSG_ALLOC(vn, rename);
2119 rename_msg->pvnr_cookie_src = VPTOPNC(fvp);
2120 rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp);
2121 if (tvp)
2122 rename_msg->pvnr_cookie_targ = VPTOPNC(tvp);
2123 else
2124 rename_msg->pvnr_cookie_targ = NULL;
2125 puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred,
2126 ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp));
2127 puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred,
2128 ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp));
2129 puffs_msg_setinfo(park_rename, PUFFSOP_VN,
2130 PUFFS_VN_RENAME, VPTOPNC(fdvp));
2131
2132 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error);
2133 doabort = false;
2134 PUFFS_MSG_RELEASE(rename);
2135 error = checkerr(pmp, error, __func__);
2136
2137 /*
2138 * XXX: stay in touch with the cache. I don't like this, but
2139 * don't have a better solution either. See also puffs_link().
2140 */
2141 if (error == 0) {
2142 puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0);
2143
2144 if (PUFFS_USE_DOTDOTCACHE(pmp) &&
2145 (VPTOPP(fvp)->pn_parent != tdvp))
2146 update_parent(fvp, tdvp);
2147 }
2148
2149
2150 out:
2151 if (doabort)
2152 VOP_ABORTOP(tdvp, ap->a_tcnp);
2153 if (tvp != NULL)
2154 vput(tvp);
2155 if (tdvp == tvp)
2156 vrele(tdvp);
2157 else
2158 vput(tdvp);
2159
2160 if (doabort)
2161 VOP_ABORTOP(fdvp, ap->a_fcnp);
2162 vrele(fdvp);
2163 vrele(fvp);
2164
2165 return error;
2166 }
2167
2168 #define RWARGS(cont, iofl, move, offset, creds) \
2169 (cont)->pvnr_ioflag = (iofl); \
2170 (cont)->pvnr_resid = (move); \
2171 (cont)->pvnr_offset = (offset); \
2172 puffs_credcvt(&(cont)->pvnr_cred, creds)
2173
2174 int
2175 puffs_vnop_read(void *v)
2176 {
2177 struct vop_read_args /* {
2178 const struct vnodeop_desc *a_desc;
2179 struct vnode *a_vp;
2180 struct uio *a_uio;
2181 int a_ioflag;
2182 kauth_cred_t a_cred;
2183 } */ *ap = v;
2184 PUFFS_MSG_VARS(vn, read);
2185 struct vnode *vp = ap->a_vp;
2186 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2187 struct uio *uio = ap->a_uio;
2188 size_t tomove, argsize;
2189 vsize_t bytelen;
2190 int error;
2191
2192 read_msg = NULL;
2193 error = 0;
2194
2195 /* std sanity */
2196 if (uio->uio_resid == 0)
2197 return 0;
2198 if (uio->uio_offset < 0)
2199 return EINVAL;
2200
2201 if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
2202 const int advice = IO_ADV_DECODE(ap->a_ioflag);
2203
2204 while (uio->uio_resid > 0) {
2205 if (vp->v_size <= uio->uio_offset) {
2206 break;
2207 }
2208 bytelen = MIN(uio->uio_resid,
2209 vp->v_size - uio->uio_offset);
2210 if (bytelen == 0)
2211 break;
2212
2213 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
2214 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
2215 if (error)
2216 break;
2217 }
2218
2219 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
2220 puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0);
2221 } else {
2222 /*
2223 * in case it's not a regular file or we're operating
2224 * uncached, do read in the old-fashioned style,
2225 * i.e. explicit read operations
2226 */
2227
2228 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2229 argsize = sizeof(struct puffs_vnmsg_read);
2230 puffs_msgmem_alloc(argsize + tomove, &park_read,
2231 (void *)&read_msg, 1);
2232
2233 error = 0;
2234 while (uio->uio_resid > 0) {
2235 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2236 memset(read_msg, 0, argsize); /* XXX: touser KASSERT */
2237 RWARGS(read_msg, ap->a_ioflag, tomove,
2238 uio->uio_offset, ap->a_cred);
2239 puffs_msg_setinfo(park_read, PUFFSOP_VN,
2240 PUFFS_VN_READ, VPTOPNC(vp));
2241 puffs_msg_setdelta(park_read, tomove);
2242
2243 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data,
2244 NULL, error);
2245 error = checkerr(pmp, error, __func__);
2246 if (error)
2247 break;
2248
2249 if (read_msg->pvnr_resid > tomove) {
2250 puffs_senderr(pmp, PUFFS_ERR_READ,
2251 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2252 error = EPROTO;
2253 break;
2254 }
2255
2256 error = uiomove(read_msg->pvnr_data,
2257 tomove - read_msg->pvnr_resid, uio);
2258
2259 /*
2260 * in case the file is out of juice, resid from
2261 * userspace is != 0. and the error-case is
2262 * quite obvious
2263 */
2264 if (error || read_msg->pvnr_resid)
2265 break;
2266 }
2267
2268 puffs_msgmem_release(park_read);
2269 }
2270
2271 return error;
2272 }
2273
2274 /*
2275 * XXX: in case of a failure, this leaves uio in a bad state.
2276 * We could theoretically copy the uio and iovecs and "replay"
2277 * them the right amount after the userspace trip, but don't
2278 * bother for now.
2279 */
2280 int
2281 puffs_vnop_write(void *v)
2282 {
2283 struct vop_write_args /* {
2284 const struct vnodeop_desc *a_desc;
2285 struct vnode *a_vp;
2286 struct uio *a_uio;
2287 int a_ioflag;
2288 kauth_cred_t a_cred;
2289 } */ *ap = v;
2290 PUFFS_MSG_VARS(vn, write);
2291 struct vnode *vp = ap->a_vp;
2292 struct puffs_node *pn = VPTOPP(vp);
2293 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2294 struct uio *uio = ap->a_uio;
2295 size_t tomove, argsize;
2296 off_t oldoff, newoff, origoff;
2297 vsize_t bytelen;
2298 int error, uflags;
2299 int ubcflags;
2300
2301 error = uflags = 0;
2302 write_msg = NULL;
2303
2304 mutex_enter(&pn->pn_sizemtx);
2305
2306 if (vp->v_type == VREG && PUFFS_USE_PAGECACHE(pmp)) {
2307 ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp);
2308
2309 /*
2310 * userspace *should* be allowed to control this,
2311 * but with UBC it's a bit unclear how to handle it
2312 */
2313 if (ap->a_ioflag & IO_APPEND)
2314 uio->uio_offset = vp->v_size;
2315
2316 origoff = uio->uio_offset;
2317 while (uio->uio_resid > 0) {
2318 if (vp->v_mount->mnt_flag & MNT_RELATIME)
2319 uflags |= PUFFS_UPDATEATIME;
2320 uflags |= PUFFS_UPDATECTIME;
2321 uflags |= PUFFS_UPDATEMTIME;
2322 oldoff = uio->uio_offset;
2323 bytelen = uio->uio_resid;
2324
2325 newoff = oldoff + bytelen;
2326 if (vp->v_size < newoff) {
2327 uvm_vnp_setwritesize(vp, newoff);
2328 }
2329 error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
2330 UVM_ADV_RANDOM, ubcflags);
2331
2332 /*
2333 * In case of a ubc_uiomove() error,
2334 * opt to not extend the file at all and
2335 * return an error. Otherwise, if we attempt
2336 * to clear the memory we couldn't fault to,
2337 * we might generate a kernel page fault.
2338 */
2339 if (vp->v_size < newoff) {
2340 if (error == 0) {
2341 uflags |= PUFFS_UPDATESIZE;
2342 uvm_vnp_setsize(vp, newoff);
2343 } else {
2344 uvm_vnp_setwritesize(vp, vp->v_size);
2345 }
2346 }
2347 if (error)
2348 break;
2349
2350 /*
2351 * If we're writing large files, flush to file server
2352 * every 64k. Otherwise we can very easily exhaust
2353 * kernel and user memory, as the file server cannot
2354 * really keep up with our writing speed.
2355 *
2356 * Note: this does *NOT* honor MNT_ASYNC, because
2357 * that gives userland too much say in the kernel.
2358 */
2359 if (oldoff >> 16 != uio->uio_offset >> 16) {
2360 mutex_enter(vp->v_interlock);
2361 error = VOP_PUTPAGES(vp, oldoff & ~0xffff,
2362 uio->uio_offset & ~0xffff,
2363 PGO_CLEANIT | PGO_SYNCIO);
2364 if (error)
2365 break;
2366 }
2367 }
2368
2369 /* synchronous I/O? */
2370 if (error == 0 && ap->a_ioflag & IO_SYNC) {
2371 mutex_enter(vp->v_interlock);
2372 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2373 round_page(uio->uio_offset),
2374 PGO_CLEANIT | PGO_SYNCIO);
2375
2376 /* write through page cache? */
2377 } else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) {
2378 mutex_enter(vp->v_interlock);
2379 error = VOP_PUTPAGES(vp, trunc_page(origoff),
2380 round_page(uio->uio_offset), PGO_CLEANIT);
2381 }
2382
2383 puffs_updatenode(VPTOPP(vp), uflags, vp->v_size);
2384 } else {
2385 /* tomove is non-increasing */
2386 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2387 argsize = sizeof(struct puffs_vnmsg_write) + tomove;
2388 puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1);
2389
2390 while (uio->uio_resid > 0) {
2391 /* move data to buffer */
2392 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp);
2393 memset(write_msg, 0, argsize); /* XXX: touser KASSERT */
2394 RWARGS(write_msg, ap->a_ioflag, tomove,
2395 uio->uio_offset, ap->a_cred);
2396 error = uiomove(write_msg->pvnr_data, tomove, uio);
2397 if (error)
2398 break;
2399
2400 /* move buffer to userspace */
2401 puffs_msg_setinfo(park_write, PUFFSOP_VN,
2402 PUFFS_VN_WRITE, VPTOPNC(vp));
2403 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data,
2404 NULL, error);
2405 error = checkerr(pmp, error, __func__);
2406 if (error)
2407 break;
2408
2409 if (write_msg->pvnr_resid > tomove) {
2410 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2411 E2BIG, "resid grew", VPTOPNC(ap->a_vp));
2412 error = EPROTO;
2413 break;
2414 }
2415
2416 /* adjust file size */
2417 if (vp->v_size < uio->uio_offset)
2418 uvm_vnp_setsize(vp, uio->uio_offset);
2419
2420 /* didn't move everything? bad userspace. bail */
2421 if (write_msg->pvnr_resid != 0) {
2422 error = EIO;
2423 break;
2424 }
2425 }
2426 puffs_msgmem_release(park_write);
2427 }
2428
2429 mutex_exit(&pn->pn_sizemtx);
2430 return error;
2431 }
2432
2433 int
2434 puffs_vnop_print(void *v)
2435 {
2436 struct vop_print_args /* {
2437 struct vnode *a_vp;
2438 } */ *ap = v;
2439 PUFFS_MSG_VARS(vn, print);
2440 struct vnode *vp = ap->a_vp;
2441 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2442 struct puffs_node *pn = vp->v_data;
2443 int error;
2444
2445 /* kernel portion */
2446 printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n"
2447 "\tuserspace cookie: %p", vp, pn, pn->pn_cookie);
2448 if (vp->v_type == VFIFO)
2449 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
2450 printf("\n");
2451
2452 /* userspace portion */
2453 if (EXISTSOP(pmp, PRINT)) {
2454 PUFFS_MSG_ALLOC(vn, print);
2455 puffs_msg_setinfo(park_print, PUFFSOP_VN,
2456 PUFFS_VN_PRINT, VPTOPNC(vp));
2457 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_print, vp->v_data,
2458 NULL, error);
2459 PUFFS_MSG_RELEASE(print);
2460 }
2461
2462 return 0;
2463 }
2464
2465 int
2466 puffs_vnop_pathconf(void *v)
2467 {
2468 struct vop_pathconf_args /* {
2469 const struct vnodeop_desc *a_desc;
2470 struct vnode *a_vp;
2471 int a_name;
2472 register_t *a_retval;
2473 } */ *ap = v;
2474 PUFFS_MSG_VARS(vn, pathconf);
2475 struct vnode *vp = ap->a_vp;
2476 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2477 int error;
2478
2479 PUFFS_MSG_ALLOC(vn, pathconf);
2480 pathconf_msg->pvnr_name = ap->a_name;
2481 puffs_msg_setinfo(park_pathconf, PUFFSOP_VN,
2482 PUFFS_VN_PATHCONF, VPTOPNC(vp));
2483 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error);
2484 error = checkerr(pmp, error, __func__);
2485 if (!error)
2486 *ap->a_retval = pathconf_msg->pvnr_retval;
2487 PUFFS_MSG_RELEASE(pathconf);
2488
2489 return error;
2490 }
2491
2492 int
2493 puffs_vnop_advlock(void *v)
2494 {
2495 struct vop_advlock_args /* {
2496 const struct vnodeop_desc *a_desc;
2497 struct vnode *a_vp;
2498 void *a_id;
2499 int a_op;
2500 struct flock *a_fl;
2501 int a_flags;
2502 } */ *ap = v;
2503 PUFFS_MSG_VARS(vn, advlock);
2504 struct vnode *vp = ap->a_vp;
2505 struct puffs_node *pn = VPTOPP(vp);
2506 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2507 int error;
2508
2509 if (!EXISTSOP(pmp, ADVLOCK))
2510 return lf_advlock(ap, &pn->pn_lockf, vp->v_size);
2511
2512 PUFFS_MSG_ALLOC(vn, advlock);
2513 (void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl,
2514 sizeof(advlock_msg->pvnr_fl));
2515 advlock_msg->pvnr_id = ap->a_id;
2516 advlock_msg->pvnr_op = ap->a_op;
2517 advlock_msg->pvnr_flags = ap->a_flags;
2518 puffs_msg_setinfo(park_advlock, PUFFSOP_VN,
2519 PUFFS_VN_ADVLOCK, VPTOPNC(vp));
2520 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error);
2521 error = checkerr(pmp, error, __func__);
2522 PUFFS_MSG_RELEASE(advlock);
2523
2524 return error;
2525 }
2526
2527 int
2528 puffs_vnop_abortop(void *v)
2529 {
2530 struct vop_abortop_args /* {
2531 struct vnode *a_dvp;
2532 struct componentname *a_cnp;
2533 }; */ *ap = v;
2534 PUFFS_MSG_VARS(vn, abortop);
2535 struct vnode *dvp = ap->a_dvp;
2536 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount);
2537 struct componentname *cnp = ap->a_cnp;
2538
2539 if (EXISTSOP(pmp, ABORTOP)) {
2540 PUFFS_MSG_ALLOC(vn, abortop);
2541 puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred,
2542 cnp, PUFFS_USE_FULLPNBUF(pmp));
2543 puffs_msg_setfaf(park_abortop);
2544 puffs_msg_setinfo(park_abortop, PUFFSOP_VN,
2545 PUFFS_VN_ABORTOP, VPTOPNC(dvp));
2546
2547 puffs_msg_enqueue(pmp, park_abortop);
2548 PUFFS_MSG_RELEASE(abortop);
2549 }
2550
2551 return genfs_abortop(v);
2552 }
2553
2554 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC)
2555
2556 /*
2557 * This maps itself to PUFFS_VN_READ/WRITE for data transfer.
2558 */
2559 int
2560 puffs_vnop_strategy(void *v)
2561 {
2562 struct vop_strategy_args /* {
2563 const struct vnodeop_desc *a_desc;
2564 struct vnode *a_vp;
2565 struct buf *a_bp;
2566 } */ *ap = v;
2567 PUFFS_MSG_VARS(vn, rw);
2568 struct vnode *vp = ap->a_vp;
2569 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2570 struct puffs_node *pn;
2571 struct buf *bp;
2572 size_t argsize;
2573 size_t tomove, moved;
2574 int error, dofaf, cansleep, dobiodone;
2575
2576 pmp = MPTOPUFFSMP(vp->v_mount);
2577 bp = ap->a_bp;
2578 error = 0;
2579 dofaf = 0;
2580 cansleep = 0;
2581 pn = VPTOPP(vp);
2582 park_rw = NULL; /* explicit */
2583 dobiodone = 1;
2584
2585 if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ))
2586 || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE)))
2587 ERROUT(EOPNOTSUPP);
2588
2589 /*
2590 * Short-circuit optimization: don't flush buffer in between
2591 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references.
2592 */
2593 if (pn->pn_stat & PNODE_DYING) {
2594 KASSERT(BUF_ISWRITE(bp));
2595 bp->b_resid = 0;
2596 goto out;
2597 }
2598
2599 #ifdef DIAGNOSTIC
2600 if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX)
2601 panic("puffs_strategy: wildly inappropriate buf bcount %d",
2602 bp->b_bcount);
2603 #endif
2604
2605 /*
2606 * See explanation for the necessity of a FAF in puffs_fsync.
2607 *
2608 * Also, do FAF in case we're suspending.
2609 * See puffs_vfsops.c:pageflush()
2610 */
2611 if (BUF_ISWRITE(bp)) {
2612 mutex_enter(vp->v_interlock);
2613 if (vp->v_iflag & VI_XLOCK)
2614 dofaf = 1;
2615 if (pn->pn_stat & PNODE_FAF)
2616 dofaf = 1;
2617 mutex_exit(vp->v_interlock);
2618 }
2619
2620 cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1;
2621
2622 KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp));
2623
2624 /* allocate transport structure */
2625 tomove = PUFFS_TOMOVE(bp->b_bcount, pmp);
2626 argsize = sizeof(struct puffs_vnmsg_rw);
2627 error = puffs_msgmem_alloc(argsize + tomove, &park_rw,
2628 (void *)&rw_msg, cansleep);
2629 if (error)
2630 goto out;
2631 RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED);
2632
2633 /* 2x2 cases: read/write, faf/nofaf */
2634 if (BUF_ISREAD(bp)) {
2635 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2636 PUFFS_VN_READ, VPTOPNC(vp));
2637 puffs_msg_setdelta(park_rw, tomove);
2638 if (BIOASYNC(bp)) {
2639 puffs_msg_setcall(park_rw,
2640 puffs_parkdone_asyncbioread, bp);
2641 puffs_msg_enqueue(pmp, park_rw);
2642 dobiodone = 0;
2643 } else {
2644 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data,
2645 NULL, error);
2646 error = checkerr(pmp, error, __func__);
2647 if (error)
2648 goto out;
2649
2650 if (rw_msg->pvnr_resid > tomove) {
2651 puffs_senderr(pmp, PUFFS_ERR_READ,
2652 E2BIG, "resid grew", VPTOPNC(vp));
2653 ERROUT(EPROTO);
2654 }
2655
2656 moved = tomove - rw_msg->pvnr_resid;
2657
2658 (void)memcpy(bp->b_data, rw_msg->pvnr_data, moved);
2659 bp->b_resid = bp->b_bcount - moved;
2660 }
2661 } else {
2662 puffs_msg_setinfo(park_rw, PUFFSOP_VN,
2663 PUFFS_VN_WRITE, VPTOPNC(vp));
2664 /*
2665 * make pages read-only before we write them if we want
2666 * write caching info
2667 */
2668 if (PUFFS_WCACHEINFO(pmp)) {
2669 struct uvm_object *uobj = &vp->v_uobj;
2670 int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT;
2671 struct vm_page *vmp;
2672 int i;
2673
2674 for (i = 0; i < npages; i++) {
2675 vmp= uvm_pageratop((vaddr_t)bp->b_data
2676 + (i << PAGE_SHIFT));
2677 DPRINTF(("puffs_strategy: write-protecting "
2678 "vp %p page %p, offset %" PRId64"\n",
2679 vp, vmp, vmp->offset));
2680 mutex_enter(uobj->vmobjlock);
2681 vmp->flags |= PG_RDONLY;
2682 pmap_page_protect(vmp, VM_PROT_READ);
2683 mutex_exit(uobj->vmobjlock);
2684 }
2685 }
2686
2687 (void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove);
2688 if (dofaf) {
2689 puffs_msg_setfaf(park_rw);
2690 } else if (BIOASYNC(bp)) {
2691 puffs_msg_setcall(park_rw,
2692 puffs_parkdone_asyncbiowrite, bp);
2693 dobiodone = 0;
2694 }
2695
2696 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error);
2697
2698 if (dobiodone == 0)
2699 goto out;
2700
2701 /*
2702 * XXXXXXXX: wrong, but kernel can't survive strategy
2703 * failure currently. Here, have one more X: X.
2704 */
2705 if (error != ENOMEM)
2706 error = 0;
2707
2708 error = checkerr(pmp, error, __func__);
2709 if (error)
2710 goto out;
2711
2712 if (rw_msg->pvnr_resid > tomove) {
2713 puffs_senderr(pmp, PUFFS_ERR_WRITE,
2714 E2BIG, "resid grew", VPTOPNC(vp));
2715 ERROUT(EPROTO);
2716 }
2717
2718 /*
2719 * FAF moved everything. Frankly, we don't
2720 * really have a choice.
2721 */
2722 if (dofaf && error == 0)
2723 moved = tomove;
2724 else
2725 moved = tomove - rw_msg->pvnr_resid;
2726
2727 bp->b_resid = bp->b_bcount - moved;
2728 if (bp->b_resid != 0) {
2729 ERROUT(EIO);
2730 }
2731 }
2732
2733 out:
2734 if (park_rw)
2735 puffs_msgmem_release(park_rw);
2736
2737 if (error)
2738 bp->b_error = error;
2739
2740 if (error || dobiodone)
2741 biodone(bp);
2742
2743 return error;
2744 }
2745
2746 int
2747 puffs_vnop_mmap(void *v)
2748 {
2749 struct vop_mmap_args /* {
2750 const struct vnodeop_desc *a_desc;
2751 struct vnode *a_vp;
2752 vm_prot_t a_prot;
2753 kauth_cred_t a_cred;
2754 } */ *ap = v;
2755 PUFFS_MSG_VARS(vn, mmap);
2756 struct vnode *vp = ap->a_vp;
2757 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2758 int error;
2759
2760 if (!PUFFS_USE_PAGECACHE(pmp))
2761 return genfs_eopnotsupp(v);
2762
2763 if (EXISTSOP(pmp, MMAP)) {
2764 PUFFS_MSG_ALLOC(vn, mmap);
2765 mmap_msg->pvnr_prot = ap->a_prot;
2766 puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred);
2767 puffs_msg_setinfo(park_mmap, PUFFSOP_VN,
2768 PUFFS_VN_MMAP, VPTOPNC(vp));
2769
2770 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error);
2771 error = checkerr(pmp, error, __func__);
2772 PUFFS_MSG_RELEASE(mmap);
2773 } else {
2774 error = genfs_mmap(v);
2775 }
2776
2777 return error;
2778 }
2779
2780
2781 /*
2782 * The rest don't get a free trip to userspace and back, they
2783 * have to stay within the kernel.
2784 */
2785
2786 /*
2787 * bmap doesn't really make any sense for puffs, so just 1:1 map it.
2788 * well, maybe somehow, somewhere, some day ....
2789 */
2790 int
2791 puffs_vnop_bmap(void *v)
2792 {
2793 struct vop_bmap_args /* {
2794 const struct vnodeop_desc *a_desc;
2795 struct vnode *a_vp;
2796 daddr_t a_bn;
2797 struct vnode **a_vpp;
2798 daddr_t *a_bnp;
2799 int *a_runp;
2800 } */ *ap = v;
2801 struct puffs_mount *pmp;
2802
2803 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2804
2805 if (ap->a_vpp)
2806 *ap->a_vpp = ap->a_vp;
2807 if (ap->a_bnp)
2808 *ap->a_bnp = ap->a_bn;
2809 if (ap->a_runp)
2810 *ap->a_runp
2811 = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1;
2812
2813 return 0;
2814 }
2815
2816 /*
2817 * Handle getpages faults in puffs. We let genfs_getpages() do most
2818 * of the dirty work, but we come in this route to do accounting tasks.
2819 * If the user server has specified functions for cache notifications
2820 * about reads and/or writes, we record which type of operation we got,
2821 * for which page range, and proceed to issue a FAF notification to the
2822 * server about it.
2823 */
2824 int
2825 puffs_vnop_getpages(void *v)
2826 {
2827 struct vop_getpages_args /* {
2828 const struct vnodeop_desc *a_desc;
2829 struct vnode *a_vp;
2830 voff_t a_offset;
2831 struct vm_page **a_m;
2832 int *a_count;
2833 int a_centeridx;
2834 vm_prot_t a_access_type;
2835 int a_advice;
2836 int a_flags;
2837 } */ *ap = v;
2838 struct puffs_mount *pmp;
2839 struct puffs_node *pn;
2840 struct vnode *vp;
2841 struct vm_page **pgs;
2842 struct puffs_cacheinfo *pcinfo = NULL;
2843 struct puffs_cacherun *pcrun;
2844 void *parkmem = NULL;
2845 size_t runsizes;
2846 int i, npages, si, streakon;
2847 int error, locked, write;
2848
2849 pmp = MPTOPUFFSMP(ap->a_vp->v_mount);
2850 npages = *ap->a_count;
2851 pgs = ap->a_m;
2852 vp = ap->a_vp;
2853 pn = vp->v_data;
2854 locked = (ap->a_flags & PGO_LOCKED) != 0;
2855 write = (ap->a_access_type & VM_PROT_WRITE) != 0;
2856
2857 /* ccg xnaht - gets Wuninitialized wrong */
2858 pcrun = NULL;
2859 runsizes = 0;
2860
2861 /*
2862 * Check that we aren't trying to fault in pages which our file
2863 * server doesn't know about. This happens if we extend a file by
2864 * skipping some pages and later try to fault in pages which
2865 * are between pn_serversize and vp_size. This check optimizes
2866 * away the common case where a file is being extended.
2867 */
2868 if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) {
2869 struct vattr va;
2870
2871 /* try again later when we can block */
2872 if (locked)
2873 ERROUT(EBUSY);
2874
2875 mutex_exit(vp->v_interlock);
2876 vattr_null(&va);
2877 va.va_size = vp->v_size;
2878 error = dosetattr(vp, &va, FSCRED, 0);
2879 if (error)
2880 ERROUT(error);
2881 mutex_enter(vp->v_interlock);
2882 }
2883
2884 if (write && PUFFS_WCACHEINFO(pmp)) {
2885 #ifdef notnowjohn
2886 /* allocate worst-case memory */
2887 runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun);
2888 KASSERT(curlwp != uvm.pagedaemon_lwp || locked);
2889 pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize,
2890 locked ? KM_NOSLEEP : KM_SLEEP);
2891
2892 /*
2893 * can't block if we're locked and can't mess up caching
2894 * information for fs server. so come back later, please
2895 */
2896 if (pcinfo == NULL)
2897 ERROUT(ENOMEM);
2898
2899 parkmem = puffs_park_alloc(locked == 0);
2900 if (parkmem == NULL)
2901 ERROUT(ENOMEM);
2902
2903 pcrun = pcinfo->pcache_runs;
2904 #else
2905 (void)parkmem;
2906 #endif
2907 }
2908
2909 error = genfs_getpages(v);
2910 if (error)
2911 goto out;
2912
2913 if (PUFFS_WCACHEINFO(pmp) == 0)
2914 goto out;
2915
2916 /*
2917 * Let's see whose fault it was and inform the user server of
2918 * possibly read/written pages. Map pages from read faults
2919 * strictly read-only, since otherwise we might miss info on
2920 * when the page is actually write-faulted to.
2921 */
2922 if (!locked)
2923 mutex_enter(vp->v_uobj.vmobjlock);
2924 for (i = 0, si = 0, streakon = 0; i < npages; i++) {
2925 if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) {
2926 if (streakon && write) {
2927 streakon = 0;
2928 pcrun[si].pcache_runend
2929 = trunc_page(pgs[i]->offset) + PAGE_MASK;
2930 si++;
2931 }
2932 continue;
2933 }
2934 if (streakon == 0 && write) {
2935 streakon = 1;
2936 pcrun[si].pcache_runstart = pgs[i]->offset;
2937 }
2938
2939 if (!write)
2940 pgs[i]->flags |= PG_RDONLY;
2941 }
2942 /* was the last page part of our streak? */
2943 if (streakon) {
2944 pcrun[si].pcache_runend
2945 = trunc_page(pgs[i-1]->offset) + PAGE_MASK;
2946 si++;
2947 }
2948 if (!locked)
2949 mutex_exit(vp->v_uobj.vmobjlock);
2950
2951 KASSERT(si <= (npages / 2) + 1);
2952
2953 #ifdef notnowjohn
2954 /* send results to userspace */
2955 if (write)
2956 puffs_cacheop(pmp, parkmem, pcinfo,
2957 sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp));
2958 #endif
2959
2960 out:
2961 if (error) {
2962 if (pcinfo != NULL)
2963 kmem_free(pcinfo,
2964 sizeof(struct puffs_cacheinfo) + runsizes);
2965 #ifdef notnowjohn
2966 if (parkmem != NULL)
2967 puffs_park_release(parkmem, 1);
2968 #endif
2969 }
2970
2971 return error;
2972 }
2973
2974 /*
2975 * Extended attribute support.
2976 */
2977
2978 int
2979 puffs_vnop_getextattr(void *v)
2980 {
2981 struct vop_getextattr_args /*
2982 struct vnode *a_vp;
2983 int a_attrnamespace;
2984 const char *a_name;
2985 struct uio *a_uio;
2986 size_t *a_size;
2987 kauth_cred_t a_cred;
2988 }; */ *ap = v;
2989 PUFFS_MSG_VARS(vn, getextattr);
2990 struct vnode *vp = ap->a_vp;
2991 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
2992 int attrnamespace = ap->a_attrnamespace;
2993 const char *name = ap->a_name;
2994 struct uio *uio = ap->a_uio;
2995 size_t *sizep = ap->a_size;
2996 size_t tomove, resid;
2997 int error;
2998
2999 if (uio)
3000 resid = uio->uio_resid;
3001 else
3002 resid = 0;
3003
3004 tomove = PUFFS_TOMOVE(resid, pmp);
3005 if (tomove != resid) {
3006 error = E2BIG;
3007 goto out;
3008 }
3009
3010 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove,
3011 &park_getextattr, (void *)&getextattr_msg, 1);
3012
3013 getextattr_msg->pvnr_attrnamespace = attrnamespace;
3014 strlcpy(getextattr_msg->pvnr_attrname, name,
3015 sizeof(getextattr_msg->pvnr_attrname));
3016 puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred);
3017 if (sizep)
3018 getextattr_msg->pvnr_datasize = 1;
3019 getextattr_msg->pvnr_resid = tomove;
3020
3021 puffs_msg_setinfo(park_getextattr,
3022 PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp));
3023 puffs_msg_setdelta(park_getextattr, tomove);
3024 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error);
3025
3026 error = checkerr(pmp, error, __func__);
3027 if (error)
3028 goto out;
3029
3030 resid = getextattr_msg->pvnr_resid;
3031 if (resid > tomove) {
3032 puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG,
3033 "resid grew", VPTOPNC(vp));
3034 error = EPROTO;
3035 goto out;
3036 }
3037
3038 if (sizep)
3039 *sizep = getextattr_msg->pvnr_datasize;
3040 if (uio)
3041 error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio);
3042
3043 out:
3044 PUFFS_MSG_RELEASE(getextattr);
3045 return error;
3046 }
3047
3048 int
3049 puffs_vnop_setextattr(void *v)
3050 {
3051 struct vop_setextattr_args /* {
3052 struct vnode *a_vp;
3053 int a_attrnamespace;
3054 const char *a_name;
3055 struct uio *a_uio;
3056 kauth_cred_t a_cred;
3057 }; */ *ap = v;
3058 PUFFS_MSG_VARS(vn, setextattr);
3059 struct vnode *vp = ap->a_vp;
3060 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3061 int attrnamespace = ap->a_attrnamespace;
3062 const char *name = ap->a_name;
3063 struct uio *uio = ap->a_uio;
3064 size_t tomove, resid;
3065 int error;
3066
3067 if (uio)
3068 resid = uio->uio_resid;
3069 else
3070 resid = 0;
3071
3072 tomove = PUFFS_TOMOVE(resid, pmp);
3073 if (tomove != resid) {
3074 error = E2BIG;
3075 goto out;
3076 }
3077
3078 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove,
3079 &park_setextattr, (void *)&setextattr_msg, 1);
3080
3081 setextattr_msg->pvnr_attrnamespace = attrnamespace;
3082 strlcpy(setextattr_msg->pvnr_attrname, name,
3083 sizeof(setextattr_msg->pvnr_attrname));
3084 puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred);
3085 setextattr_msg->pvnr_resid = tomove;
3086
3087 if (uio) {
3088 error = uiomove(setextattr_msg->pvnr_data, tomove, uio);
3089 if (error)
3090 goto out;
3091 }
3092
3093 puffs_msg_setinfo(park_setextattr,
3094 PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp));
3095 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error);
3096
3097 error = checkerr(pmp, error, __func__);
3098 if (error)
3099 goto out;
3100
3101 if (setextattr_msg->pvnr_resid != 0)
3102 error = EIO;
3103
3104 out:
3105 PUFFS_MSG_RELEASE(setextattr);
3106
3107 return error;
3108 }
3109
3110 int
3111 puffs_vnop_listextattr(void *v)
3112 {
3113 struct vop_listextattr_args /* {
3114 struct vnode *a_vp;
3115 int a_attrnamespace;
3116 struct uio *a_uio;
3117 size_t *a_size;
3118 int a_flag,
3119 kauth_cred_t a_cred;
3120 }; */ *ap = v;
3121 PUFFS_MSG_VARS(vn, listextattr);
3122 struct vnode *vp = ap->a_vp;
3123 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3124 int attrnamespace = ap->a_attrnamespace;
3125 struct uio *uio = ap->a_uio;
3126 size_t *sizep = ap->a_size;
3127 int flag = ap->a_flag;
3128 size_t tomove, resid;
3129 int error;
3130
3131 if (uio)
3132 resid = uio->uio_resid;
3133 else
3134 resid = 0;
3135
3136 tomove = PUFFS_TOMOVE(resid, pmp);
3137 if (tomove != resid) {
3138 error = E2BIG;
3139 goto out;
3140 }
3141
3142 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_listextattr) + tomove,
3143 &park_listextattr, (void *)&listextattr_msg, 1);
3144
3145 listextattr_msg->pvnr_attrnamespace = attrnamespace;
3146 listextattr_msg->pvnr_flag = flag;
3147 puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred);
3148 listextattr_msg->pvnr_resid = tomove;
3149 if (sizep)
3150 listextattr_msg->pvnr_datasize = 1;
3151
3152 puffs_msg_setinfo(park_listextattr,
3153 PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp));
3154 puffs_msg_setdelta(park_listextattr, tomove);
3155 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error);
3156
3157 error = checkerr(pmp, error, __func__);
3158 if (error)
3159 goto out;
3160
3161 resid = listextattr_msg->pvnr_resid;
3162 if (resid > tomove) {
3163 puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG,
3164 "resid grew", VPTOPNC(vp));
3165 error = EPROTO;
3166 goto out;
3167 }
3168
3169 if (sizep)
3170 *sizep = listextattr_msg->pvnr_datasize;
3171 if (uio)
3172 error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio);
3173
3174 out:
3175 PUFFS_MSG_RELEASE(listextattr);
3176 return error;
3177 }
3178
3179 int
3180 puffs_vnop_deleteextattr(void *v)
3181 {
3182 struct vop_deleteextattr_args /* {
3183 struct vnode *a_vp;
3184 int a_attrnamespace;
3185 const char *a_name;
3186 kauth_cred_t a_cred;
3187 }; */ *ap = v;
3188 PUFFS_MSG_VARS(vn, deleteextattr);
3189 struct vnode *vp = ap->a_vp;
3190 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount);
3191 int attrnamespace = ap->a_attrnamespace;
3192 const char *name = ap->a_name;
3193 int error;
3194
3195 PUFFS_MSG_ALLOC(vn, deleteextattr);
3196 deleteextattr_msg->pvnr_attrnamespace = attrnamespace;
3197 strlcpy(deleteextattr_msg->pvnr_attrname, name,
3198 sizeof(deleteextattr_msg->pvnr_attrname));
3199 puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred);
3200
3201 puffs_msg_setinfo(park_deleteextattr,
3202 PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp));
3203 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr,
3204 vp->v_data, NULL, error);
3205
3206 error = checkerr(pmp, error, __func__);
3207
3208 PUFFS_MSG_RELEASE(deleteextattr);
3209 return error;
3210 }
3211
3212 /*
3213 * spec & fifo. These call the miscfs spec and fifo vectors, but issue
3214 * FAF update information for the puffs node first.
3215 */
3216 int
3217 puffs_vnop_spec_read(void *v)
3218 {
3219 struct vop_read_args /* {
3220 const struct vnodeop_desc *a_desc;
3221 struct vnode *a_vp;
3222 struct uio *a_uio;
3223 int a_ioflag;
3224 kauth_cred_t a_cred;
3225 } */ *ap = v;
3226
3227 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3228 return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v);
3229 }
3230
3231 int
3232 puffs_vnop_spec_write(void *v)
3233 {
3234 struct vop_write_args /* {
3235 const struct vnodeop_desc *a_desc;
3236 struct vnode *a_vp;
3237 struct uio *a_uio;
3238 int a_ioflag;
3239 kauth_cred_t a_cred;
3240 } */ *ap = v;
3241
3242 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3243 return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v);
3244 }
3245
3246 int
3247 puffs_vnop_fifo_read(void *v)
3248 {
3249 struct vop_read_args /* {
3250 const struct vnodeop_desc *a_desc;
3251 struct vnode *a_vp;
3252 struct uio *a_uio;
3253 int a_ioflag;
3254 kauth_cred_t a_cred;
3255 } */ *ap = v;
3256
3257 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0);
3258 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v);
3259 }
3260
3261 int
3262 puffs_vnop_fifo_write(void *v)
3263 {
3264 struct vop_write_args /* {
3265 const struct vnodeop_desc *a_desc;
3266 struct vnode *a_vp;
3267 struct uio *a_uio;
3268 int a_ioflag;
3269 kauth_cred_t a_cred;
3270 } */ *ap = v;
3271
3272 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0);
3273 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v);
3274 }
3275