Print this page
NEX-16625 Max amount of iSCSI targets is hard limited with doorfs core definitions
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Evan Layton <evan.layton@nexenta.com>
NEX-5164 backport illumos 6514 AS_* lock macros simplification
Reviewed by: Kevin Crowe <kevin.crowe@nexenta.com>
6514 AS_* lock macros simplification
Reviewed by: Piotr Jasiukajtis <estibi@me.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Albert Lee <trisk@omniti.com>
Approved by: Dan McDonald <danmcd@omniti.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/doorfs/door_sys.c
+++ new/usr/src/uts/common/fs/doorfs/door_sys.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2016 by Delphix. All rights reserved.
25 25 */
26 26
27 27 /*
28 28 * System call I/F to doors (outside of vnodes I/F) and misc support
29 29 * routines
30 30 */
31 31 #include <sys/types.h>
32 32 #include <sys/systm.h>
33 33 #include <sys/door.h>
34 34 #include <sys/door_data.h>
35 35 #include <sys/proc.h>
36 36 #include <sys/thread.h>
37 37 #include <sys/prsystm.h>
38 38 #include <sys/procfs.h>
39 39 #include <sys/class.h>
40 40 #include <sys/cred.h>
41 41 #include <sys/kmem.h>
42 42 #include <sys/cmn_err.h>
43 43 #include <sys/stack.h>
44 44 #include <sys/debug.h>
45 45 #include <sys/cpuvar.h>
46 46 #include <sys/file.h>
47 47 #include <sys/fcntl.h>
48 48 #include <sys/vnode.h>
49 49 #include <sys/vfs.h>
50 50 #include <sys/vfs_opreg.h>
51 51 #include <sys/sobject.h>
52 52 #include <sys/schedctl.h>
53 53 #include <sys/callb.h>
54 54 #include <sys/ucred.h>
55 55
56 56 #include <sys/mman.h>
57 57 #include <sys/sysmacros.h>
58 58 #include <sys/vmsystm.h>
59 59 #include <vm/as.h>
60 60 #include <vm/hat.h>
61 61 #include <vm/page.h>
62 62 #include <vm/seg.h>
63 63 #include <vm/seg_vn.h>
64 64 #include <vm/seg_vn.h>
65 65 #include <vm/seg_kpm.h>
66 66
67 67 #include <sys/modctl.h>
68 68 #include <sys/syscall.h>
69 69 #include <sys/pathname.h>
70 70 #include <sys/rctl.h>
71 71
72 72 /*
73 73 * The maximum amount of data (in bytes) that will be transferred using
|
↓ open down ↓ |
73 lines elided |
↑ open up ↑ |
74 74 * an intermediate kernel buffer. For sizes greater than this we map
75 75 * in the destination pages and perform a 1-copy transfer.
76 76 */
77 77 size_t door_max_arg = 16 * 1024;
78 78
79 79 /*
80 80 * Maximum amount of data that will be transferred in a reply to a
81 81 * door_upcall. Need to guard against a process returning huge amounts
82 82 * of data and getting the kernel stuck in kmem_alloc.
83 83 */
84 -size_t door_max_upcall_reply = 1024 * 1024;
84 +size_t door_max_upcall_reply = 4 * 1024 * 1024;
85 85
86 86 /*
87 87 * Maximum number of descriptors allowed to be passed in a single
88 88 * door_call or door_return. We need to allocate kernel memory
89 89 * for all of them at once, so we can't let it scale without limit.
90 90 */
91 91 uint_t door_max_desc = 1024;
92 92
93 93 /*
94 94 * Definition of a door handle, used by other kernel subsystems when
95 95 * calling door functions. This is really a file structure but we
96 96 * want to hide that fact.
97 97 */
98 98 struct __door_handle {
99 99 file_t dh_file;
100 100 };
101 101
102 102 #define DHTOF(dh) ((file_t *)(dh))
103 103 #define FTODH(fp) ((door_handle_t)(fp))
104 104
105 105 static int doorfs(long, long, long, long, long, long);
106 106
107 107 static struct sysent door_sysent = {
108 108 6,
109 109 SE_ARGC | SE_NOUNLOAD,
110 110 (int (*)())doorfs,
111 111 };
112 112
113 113 static struct modlsys modlsys = {
114 114 &mod_syscallops, "doors", &door_sysent
115 115 };
116 116
117 117 #ifdef _SYSCALL32_IMPL
118 118
119 119 static int
120 120 doorfs32(int32_t arg1, int32_t arg2, int32_t arg3, int32_t arg4,
121 121 int32_t arg5, int32_t subcode);
122 122
123 123 static struct sysent door_sysent32 = {
124 124 6,
125 125 SE_ARGC | SE_NOUNLOAD,
126 126 (int (*)())doorfs32,
127 127 };
128 128
129 129 static struct modlsys modlsys32 = {
130 130 &mod_syscallops32,
131 131 "32-bit door syscalls",
132 132 &door_sysent32
133 133 };
134 134 #endif
135 135
136 136 static struct modlinkage modlinkage = {
137 137 MODREV_1,
138 138 &modlsys,
139 139 #ifdef _SYSCALL32_IMPL
140 140 &modlsys32,
141 141 #endif
142 142 NULL
143 143 };
144 144
145 145 dev_t doordev;
146 146
147 147 extern struct vfs door_vfs;
148 148 extern struct vnodeops *door_vnodeops;
149 149
150 150 int
151 151 _init(void)
152 152 {
153 153 static const fs_operation_def_t door_vfsops_template[] = {
154 154 NULL, NULL
155 155 };
156 156 extern const fs_operation_def_t door_vnodeops_template[];
157 157 vfsops_t *door_vfsops;
158 158 major_t major;
159 159 int error;
160 160
161 161 mutex_init(&door_knob, NULL, MUTEX_DEFAULT, NULL);
162 162 if ((major = getudev()) == (major_t)-1)
163 163 return (ENXIO);
164 164 doordev = makedevice(major, 0);
165 165
166 166 /* Create a dummy vfs */
167 167 error = vfs_makefsops(door_vfsops_template, &door_vfsops);
168 168 if (error != 0) {
169 169 cmn_err(CE_WARN, "door init: bad vfs ops");
170 170 return (error);
171 171 }
172 172 VFS_INIT(&door_vfs, door_vfsops, NULL);
173 173 door_vfs.vfs_flag = VFS_RDONLY;
174 174 door_vfs.vfs_dev = doordev;
175 175 vfs_make_fsid(&(door_vfs.vfs_fsid), doordev, 0);
176 176
177 177 error = vn_make_ops("doorfs", door_vnodeops_template, &door_vnodeops);
178 178 if (error != 0) {
179 179 vfs_freevfsops(door_vfsops);
180 180 cmn_err(CE_WARN, "door init: bad vnode ops");
181 181 return (error);
182 182 }
183 183 return (mod_install(&modlinkage));
184 184 }
185 185
186 186 int
187 187 _info(struct modinfo *modinfop)
188 188 {
189 189 return (mod_info(&modlinkage, modinfop));
190 190 }
191 191
192 192 /* system call functions */
193 193 static int door_call(int, void *);
194 194 static int door_return(caddr_t, size_t, door_desc_t *, uint_t, caddr_t, size_t);
195 195 static int door_create(void (*pc_cookie)(void *, char *, size_t, door_desc_t *,
196 196 uint_t), void *data_cookie, uint_t);
197 197 static int door_revoke(int);
198 198 static int door_info(int, struct door_info *);
199 199 static int door_ucred(struct ucred_s *);
200 200 static int door_bind(int);
201 201 static int door_unbind(void);
202 202 static int door_unref(void);
203 203 static int door_getparam(int, int, size_t *);
204 204 static int door_setparam(int, int, size_t);
205 205
206 206 #define DOOR_RETURN_OLD 4 /* historic value, for s10 */
207 207
208 208 /*
209 209 * System call wrapper for all door related system calls
210 210 */
211 211 static int
212 212 doorfs(long arg1, long arg2, long arg3, long arg4, long arg5, long subcode)
213 213 {
214 214 switch (subcode) {
215 215 case DOOR_CALL:
216 216 return (door_call(arg1, (void *)arg2));
217 217 case DOOR_RETURN: {
218 218 door_return_desc_t *drdp = (door_return_desc_t *)arg3;
219 219
220 220 if (drdp != NULL) {
221 221 door_return_desc_t drd;
222 222 if (copyin(drdp, &drd, sizeof (drd)))
223 223 return (EFAULT);
224 224 return (door_return((caddr_t)arg1, arg2, drd.desc_ptr,
225 225 drd.desc_num, (caddr_t)arg4, arg5));
226 226 }
227 227 return (door_return((caddr_t)arg1, arg2, NULL,
228 228 0, (caddr_t)arg4, arg5));
229 229 }
230 230 case DOOR_RETURN_OLD:
231 231 /*
232 232 * In order to support the S10 runtime environment, we
233 233 * still respond to the old syscall subcode for door_return.
234 234 * We treat it as having no stack limits. This code should
235 235 * be removed when such support is no longer needed.
236 236 */
237 237 return (door_return((caddr_t)arg1, arg2, (door_desc_t *)arg3,
238 238 arg4, (caddr_t)arg5, 0));
239 239 case DOOR_CREATE:
240 240 return (door_create((void (*)())arg1, (void *)arg2, arg3));
241 241 case DOOR_REVOKE:
242 242 return (door_revoke(arg1));
243 243 case DOOR_INFO:
244 244 return (door_info(arg1, (struct door_info *)arg2));
245 245 case DOOR_BIND:
246 246 return (door_bind(arg1));
247 247 case DOOR_UNBIND:
248 248 return (door_unbind());
249 249 case DOOR_UNREFSYS:
250 250 return (door_unref());
251 251 case DOOR_UCRED:
252 252 return (door_ucred((struct ucred_s *)arg1));
253 253 case DOOR_GETPARAM:
254 254 return (door_getparam(arg1, arg2, (size_t *)arg3));
255 255 case DOOR_SETPARAM:
256 256 return (door_setparam(arg1, arg2, arg3));
257 257 default:
258 258 return (set_errno(EINVAL));
259 259 }
260 260 }
261 261
262 262 #ifdef _SYSCALL32_IMPL
263 263 /*
264 264 * System call wrapper for all door related system calls from 32-bit programs.
265 265 * Needed at the moment because of the casts - they undo some damage
266 266 * that truss causes (sign-extending the stack pointer) when truss'ing
267 267 * a 32-bit program using doors.
268 268 */
269 269 static int
270 270 doorfs32(int32_t arg1, int32_t arg2, int32_t arg3,
271 271 int32_t arg4, int32_t arg5, int32_t subcode)
272 272 {
273 273 switch (subcode) {
274 274 case DOOR_CALL:
275 275 return (door_call(arg1, (void *)(uintptr_t)(caddr32_t)arg2));
276 276 case DOOR_RETURN: {
277 277 door_return_desc32_t *drdp =
278 278 (door_return_desc32_t *)(uintptr_t)(caddr32_t)arg3;
279 279 if (drdp != NULL) {
280 280 door_return_desc32_t drd;
281 281 if (copyin(drdp, &drd, sizeof (drd)))
282 282 return (EFAULT);
283 283 return (door_return(
284 284 (caddr_t)(uintptr_t)(caddr32_t)arg1, arg2,
285 285 (door_desc_t *)(uintptr_t)drd.desc_ptr,
286 286 drd.desc_num, (caddr_t)(uintptr_t)(caddr32_t)arg4,
287 287 (size_t)(uintptr_t)(size32_t)arg5));
288 288 }
289 289 return (door_return((caddr_t)(uintptr_t)(caddr32_t)arg1,
290 290 arg2, NULL, 0, (caddr_t)(uintptr_t)(caddr32_t)arg4,
291 291 (size_t)(uintptr_t)(size32_t)arg5));
292 292 }
293 293 case DOOR_RETURN_OLD:
294 294 /*
295 295 * In order to support the S10 runtime environment, we
296 296 * still respond to the old syscall subcode for door_return.
297 297 * We treat it as having no stack limits. This code should
298 298 * be removed when such support is no longer needed.
299 299 */
300 300 return (door_return((caddr_t)(uintptr_t)(caddr32_t)arg1, arg2,
301 301 (door_desc_t *)(uintptr_t)(caddr32_t)arg3, arg4,
302 302 (caddr_t)(uintptr_t)(caddr32_t)arg5, 0));
303 303 case DOOR_CREATE:
304 304 return (door_create((void (*)())(uintptr_t)(caddr32_t)arg1,
305 305 (void *)(uintptr_t)(caddr32_t)arg2, arg3));
306 306 case DOOR_REVOKE:
307 307 return (door_revoke(arg1));
308 308 case DOOR_INFO:
309 309 return (door_info(arg1,
310 310 (struct door_info *)(uintptr_t)(caddr32_t)arg2));
311 311 case DOOR_BIND:
312 312 return (door_bind(arg1));
313 313 case DOOR_UNBIND:
314 314 return (door_unbind());
315 315 case DOOR_UNREFSYS:
316 316 return (door_unref());
317 317 case DOOR_UCRED:
318 318 return (door_ucred(
319 319 (struct ucred_s *)(uintptr_t)(caddr32_t)arg1));
320 320 case DOOR_GETPARAM:
321 321 return (door_getparam(arg1, arg2,
322 322 (size_t *)(uintptr_t)(caddr32_t)arg3));
323 323 case DOOR_SETPARAM:
324 324 return (door_setparam(arg1, arg2, (size_t)(size32_t)arg3));
325 325
326 326 default:
327 327 return (set_errno(EINVAL));
328 328 }
329 329 }
330 330 #endif
331 331
332 332 void shuttle_resume(kthread_t *, kmutex_t *);
333 333 void shuttle_swtch(kmutex_t *);
334 334 void shuttle_sleep(kthread_t *);
335 335
336 336 /*
337 337 * Support routines
338 338 */
339 339 static int door_create_common(void (*)(), void *, uint_t, int, int *,
340 340 file_t **);
341 341 static int door_overflow(kthread_t *, caddr_t, size_t, door_desc_t *, uint_t);
342 342 static int door_args(kthread_t *, int);
343 343 static int door_results(kthread_t *, caddr_t, size_t, door_desc_t *, uint_t);
344 344 static int door_copy(struct as *, caddr_t, caddr_t, uint_t);
345 345 static void door_server_exit(proc_t *, kthread_t *);
346 346 static void door_release_server(door_node_t *, kthread_t *);
347 347 static kthread_t *door_get_server(door_node_t *);
348 348 static door_node_t *door_lookup(int, file_t **);
349 349 static int door_translate_in(void);
350 350 static int door_translate_out(void);
351 351 static void door_fd_rele(door_desc_t *, uint_t, int);
352 352 static void door_list_insert(door_node_t *);
353 353 static void door_info_common(door_node_t *, door_info_t *, file_t *);
354 354 static int door_release_fds(door_desc_t *, uint_t);
355 355 static void door_fd_close(door_desc_t *, uint_t);
356 356 static void door_fp_close(struct file **, uint_t);
357 357
358 358 static door_data_t *
359 359 door_my_data(int create_if_missing)
360 360 {
361 361 door_data_t *ddp;
362 362
363 363 ddp = curthread->t_door;
364 364 if (create_if_missing && ddp == NULL)
365 365 ddp = curthread->t_door = kmem_zalloc(sizeof (*ddp), KM_SLEEP);
366 366
367 367 return (ddp);
368 368 }
369 369
370 370 static door_server_t *
371 371 door_my_server(int create_if_missing)
372 372 {
373 373 door_data_t *ddp = door_my_data(create_if_missing);
374 374
375 375 return ((ddp != NULL)? DOOR_SERVER(ddp) : NULL);
376 376 }
377 377
378 378 static door_client_t *
379 379 door_my_client(int create_if_missing)
380 380 {
381 381 door_data_t *ddp = door_my_data(create_if_missing);
382 382
383 383 return ((ddp != NULL)? DOOR_CLIENT(ddp) : NULL);
384 384 }
385 385
386 386 /*
387 387 * System call to create a door
388 388 */
389 389 int
390 390 door_create(void (*pc_cookie)(), void *data_cookie, uint_t attributes)
391 391 {
392 392 int fd;
393 393 int err;
394 394
395 395 if ((attributes & ~DOOR_CREATE_MASK) ||
396 396 ((attributes & (DOOR_UNREF | DOOR_UNREF_MULTI)) ==
397 397 (DOOR_UNREF | DOOR_UNREF_MULTI)))
398 398 return (set_errno(EINVAL));
399 399
400 400 if ((err = door_create_common(pc_cookie, data_cookie, attributes, 0,
401 401 &fd, NULL)) != 0)
402 402 return (set_errno(err));
403 403
404 404 f_setfd(fd, FD_CLOEXEC);
405 405 return (fd);
406 406 }
407 407
408 408 /*
409 409 * Common code for creating user and kernel doors. If a door was
410 410 * created, stores a file structure pointer in the location pointed
411 411 * to by fpp (if fpp is non-NULL) and returns 0. Also, if a non-NULL
412 412 * pointer to a file descriptor is passed in as fdp, allocates a file
413 413 * descriptor representing the door. If a door could not be created,
414 414 * returns an error.
415 415 */
416 416 static int
417 417 door_create_common(void (*pc_cookie)(), void *data_cookie, uint_t attributes,
418 418 int from_kernel, int *fdp, file_t **fpp)
419 419 {
420 420 door_node_t *dp;
421 421 vnode_t *vp;
422 422 struct file *fp;
423 423 static door_id_t index = 0;
424 424 proc_t *p = (from_kernel)? &p0 : curproc;
425 425
426 426 dp = kmem_zalloc(sizeof (door_node_t), KM_SLEEP);
427 427
428 428 dp->door_vnode = vn_alloc(KM_SLEEP);
429 429 dp->door_target = p;
430 430 dp->door_data = data_cookie;
431 431 dp->door_pc = pc_cookie;
432 432 dp->door_flags = attributes;
433 433 #ifdef _SYSCALL32_IMPL
434 434 if (!from_kernel && get_udatamodel() != DATAMODEL_NATIVE)
435 435 dp->door_data_max = UINT32_MAX;
436 436 else
437 437 #endif
438 438 dp->door_data_max = SIZE_MAX;
439 439 dp->door_data_min = 0UL;
440 440 dp->door_desc_max = (attributes & DOOR_REFUSE_DESC)? 0 : INT_MAX;
441 441
442 442 vp = DTOV(dp);
443 443 vn_setops(vp, door_vnodeops);
444 444 vp->v_type = VDOOR;
445 445 vp->v_vfsp = &door_vfs;
446 446 vp->v_data = (caddr_t)dp;
447 447 mutex_enter(&door_knob);
448 448 dp->door_index = index++;
449 449 /* add to per-process door list */
450 450 door_list_insert(dp);
451 451 mutex_exit(&door_knob);
452 452
453 453 if (falloc(vp, FREAD | FWRITE, &fp, fdp)) {
454 454 /*
455 455 * If the file table is full, remove the door from the
456 456 * per-process list, free the door, and return NULL.
457 457 */
458 458 mutex_enter(&door_knob);
459 459 door_list_delete(dp);
460 460 mutex_exit(&door_knob);
461 461 vn_free(vp);
462 462 kmem_free(dp, sizeof (door_node_t));
463 463 return (EMFILE);
464 464 }
465 465 vn_exists(vp);
466 466 if (fdp != NULL)
467 467 setf(*fdp, fp);
468 468 mutex_exit(&fp->f_tlock);
469 469
470 470 if (fpp != NULL)
471 471 *fpp = fp;
472 472 return (0);
473 473 }
474 474
475 475 static int
476 476 door_check_limits(door_node_t *dp, door_arg_t *da, int upcall)
477 477 {
478 478 ASSERT(MUTEX_HELD(&door_knob));
479 479
480 480 /* we allow unref upcalls through, despite any minimum */
481 481 if (da->data_size < dp->door_data_min &&
482 482 !(upcall && da->data_ptr == DOOR_UNREF_DATA))
483 483 return (ENOBUFS);
484 484
485 485 if (da->data_size > dp->door_data_max)
486 486 return (ENOBUFS);
487 487
488 488 if (da->desc_num > 0 && (dp->door_flags & DOOR_REFUSE_DESC))
489 489 return (ENOTSUP);
490 490
491 491 if (da->desc_num > dp->door_desc_max)
492 492 return (ENFILE);
493 493
494 494 return (0);
495 495 }
496 496
497 497 /*
498 498 * Door invocation.
499 499 */
500 500 int
501 501 door_call(int did, void *args)
502 502 {
503 503 /* Locals */
504 504 door_node_t *dp;
505 505 kthread_t *server_thread;
506 506 int error = 0;
507 507 klwp_t *lwp;
508 508 door_client_t *ct; /* curthread door_data */
509 509 door_server_t *st; /* server thread door_data */
510 510 door_desc_t *start = NULL;
511 511 uint_t ncopied = 0;
512 512 size_t dsize;
513 513 /* destructor for data returned by a kernel server */
514 514 void (*destfn)() = NULL;
515 515 void *destarg;
516 516 model_t datamodel;
517 517 int gotresults = 0;
518 518 int needcleanup = 0;
519 519 int cancel_pending;
520 520
521 521 lwp = ttolwp(curthread);
522 522 datamodel = lwp_getdatamodel(lwp);
523 523
524 524 ct = door_my_client(1);
525 525
526 526 /*
527 527 * Get the arguments
528 528 */
529 529 if (args) {
530 530 if (datamodel == DATAMODEL_NATIVE) {
531 531 if (copyin(args, &ct->d_args, sizeof (door_arg_t)) != 0)
532 532 return (set_errno(EFAULT));
533 533 } else {
534 534 door_arg32_t da32;
535 535
536 536 if (copyin(args, &da32, sizeof (door_arg32_t)) != 0)
537 537 return (set_errno(EFAULT));
538 538 ct->d_args.data_ptr =
539 539 (char *)(uintptr_t)da32.data_ptr;
540 540 ct->d_args.data_size = da32.data_size;
541 541 ct->d_args.desc_ptr =
542 542 (door_desc_t *)(uintptr_t)da32.desc_ptr;
543 543 ct->d_args.desc_num = da32.desc_num;
544 544 ct->d_args.rbuf =
545 545 (char *)(uintptr_t)da32.rbuf;
546 546 ct->d_args.rsize = da32.rsize;
547 547 }
548 548 } else {
549 549 /* No arguments, and no results allowed */
550 550 ct->d_noresults = 1;
551 551 ct->d_args.data_size = 0;
552 552 ct->d_args.desc_num = 0;
553 553 ct->d_args.rsize = 0;
554 554 }
555 555
556 556 if ((dp = door_lookup(did, NULL)) == NULL)
557 557 return (set_errno(EBADF));
558 558
559 559 /*
560 560 * We don't want to hold the door FD over the entire operation;
561 561 * instead, we put a hold on the door vnode and release the FD
562 562 * immediately
563 563 */
564 564 VN_HOLD(DTOV(dp));
565 565 releasef(did);
566 566
567 567 /*
568 568 * This should be done in shuttle_resume(), just before going to
569 569 * sleep, but we want to avoid overhead while holding door_knob.
570 570 * prstop() is just a no-op if we don't really go to sleep.
571 571 * We test not-kernel-address-space for the sake of clustering code.
572 572 */
573 573 if (lwp && lwp->lwp_nostop == 0 && curproc->p_as != &kas)
574 574 prstop(PR_REQUESTED, 0);
575 575
576 576 mutex_enter(&door_knob);
577 577 if (DOOR_INVALID(dp)) {
578 578 mutex_exit(&door_knob);
579 579 error = EBADF;
580 580 goto out;
581 581 }
582 582
583 583 /*
584 584 * before we do anything, check that we are not overflowing the
585 585 * required limits.
586 586 */
587 587 error = door_check_limits(dp, &ct->d_args, 0);
588 588 if (error != 0) {
589 589 mutex_exit(&door_knob);
590 590 goto out;
591 591 }
592 592
593 593 /*
594 594 * Check for in-kernel door server.
595 595 */
596 596 if (dp->door_target == &p0) {
597 597 caddr_t rbuf = ct->d_args.rbuf;
598 598 size_t rsize = ct->d_args.rsize;
599 599
600 600 dp->door_active++;
601 601 ct->d_kernel = 1;
602 602 ct->d_error = DOOR_WAIT;
603 603 mutex_exit(&door_knob);
604 604 /* translate file descriptors to vnodes */
605 605 if (ct->d_args.desc_num) {
606 606 error = door_translate_in();
607 607 if (error)
608 608 goto out;
609 609 }
610 610 /*
611 611 * Call kernel door server. Arguments are passed and
612 612 * returned as a door_arg pointer. When called, data_ptr
613 613 * points to user data and desc_ptr points to a kernel list
614 614 * of door descriptors that have been converted to file
615 615 * structure pointers. It's the server function's
616 616 * responsibility to copyin the data pointed to by data_ptr
617 617 * (this avoids extra copying in some cases). On return,
618 618 * data_ptr points to a user buffer of data, and desc_ptr
619 619 * points to a kernel list of door descriptors representing
620 620 * files. When a reference is passed to a kernel server,
621 621 * it is the server's responsibility to release the reference
622 622 * (by calling closef). When the server includes a
623 623 * reference in its reply, it is released as part of the
624 624 * the call (the server must duplicate the reference if
625 625 * it wants to retain a copy). The destfn, if set to
626 626 * non-NULL, is a destructor to be called when the returned
627 627 * kernel data (if any) is no longer needed (has all been
628 628 * translated and copied to user level).
629 629 */
630 630 (*(dp->door_pc))(dp->door_data, &ct->d_args,
631 631 &destfn, &destarg, &error);
632 632 mutex_enter(&door_knob);
633 633 /* not implemented yet */
634 634 if (--dp->door_active == 0 && (dp->door_flags & DOOR_DELAY))
635 635 door_deliver_unref(dp);
636 636 mutex_exit(&door_knob);
637 637 if (error)
638 638 goto out;
639 639
640 640 /* translate vnodes to files */
641 641 if (ct->d_args.desc_num) {
642 642 error = door_translate_out();
643 643 if (error)
644 644 goto out;
645 645 }
646 646 ct->d_buf = ct->d_args.rbuf;
647 647 ct->d_bufsize = ct->d_args.rsize;
648 648 if (rsize < (ct->d_args.data_size +
649 649 (ct->d_args.desc_num * sizeof (door_desc_t)))) {
650 650 /* handle overflow */
651 651 error = door_overflow(curthread, ct->d_args.data_ptr,
652 652 ct->d_args.data_size, ct->d_args.desc_ptr,
653 653 ct->d_args.desc_num);
654 654 if (error)
655 655 goto out;
656 656 /* door_overflow sets d_args rbuf and rsize */
657 657 } else {
658 658 ct->d_args.rbuf = rbuf;
659 659 ct->d_args.rsize = rsize;
660 660 }
661 661 goto results;
662 662 }
663 663
664 664 /*
665 665 * Get a server thread from the target domain
666 666 */
667 667 if ((server_thread = door_get_server(dp)) == NULL) {
668 668 if (DOOR_INVALID(dp))
669 669 error = EBADF;
670 670 else
671 671 error = EAGAIN;
672 672 mutex_exit(&door_knob);
673 673 goto out;
674 674 }
675 675
676 676 st = DOOR_SERVER(server_thread->t_door);
677 677 if (ct->d_args.desc_num || ct->d_args.data_size) {
678 678 int is_private = (dp->door_flags & DOOR_PRIVATE);
679 679 /*
680 680 * Move data from client to server
681 681 */
682 682 DOOR_T_HOLD(st);
683 683 mutex_exit(&door_knob);
684 684 error = door_args(server_thread, is_private);
685 685 mutex_enter(&door_knob);
686 686 DOOR_T_RELEASE(st);
687 687 if (error) {
688 688 /*
689 689 * We're not going to resume this thread after all
690 690 */
691 691 door_release_server(dp, server_thread);
692 692 shuttle_sleep(server_thread);
693 693 mutex_exit(&door_knob);
694 694 goto out;
695 695 }
696 696 }
697 697
698 698 dp->door_active++;
699 699 ct->d_error = DOOR_WAIT;
700 700 ct->d_args_done = 0;
701 701 st->d_caller = curthread;
702 702 st->d_active = dp;
703 703
704 704 shuttle_resume(server_thread, &door_knob);
705 705
706 706 mutex_enter(&door_knob);
707 707 shuttle_return:
708 708 if ((error = ct->d_error) < 0) { /* DOOR_WAIT or DOOR_EXIT */
709 709 /*
710 710 * Premature wakeup. Find out why (stop, forkall, sig, exit ...)
711 711 */
712 712 mutex_exit(&door_knob); /* May block in ISSIG */
713 713 cancel_pending = 0;
714 714 if (ISSIG(curthread, FORREAL) || lwp->lwp_sysabort ||
715 715 MUSTRETURN(curproc, curthread) ||
716 716 (cancel_pending = schedctl_cancel_pending()) != 0) {
717 717 /* Signal, forkall, ... */
718 718 lwp->lwp_sysabort = 0;
719 719 if (cancel_pending)
720 720 schedctl_cancel_eintr();
721 721 mutex_enter(&door_knob);
722 722 error = EINTR;
723 723 /*
724 724 * If the server has finished processing our call,
725 725 * or exited (calling door_slam()), then d_error
726 726 * will have changed. If the server hasn't finished
727 727 * yet, d_error will still be DOOR_WAIT, and we
728 728 * let it know we are not interested in any
729 729 * results by sending a SIGCANCEL, unless the door
730 730 * is marked with DOOR_NO_CANCEL.
731 731 */
732 732 if (ct->d_error == DOOR_WAIT &&
733 733 st->d_caller == curthread) {
734 734 proc_t *p = ttoproc(server_thread);
735 735
736 736 st->d_active = NULL;
737 737 st->d_caller = NULL;
738 738
739 739 if (!(dp->door_flags & DOOR_NO_CANCEL)) {
740 740 DOOR_T_HOLD(st);
741 741 mutex_exit(&door_knob);
742 742
743 743 mutex_enter(&p->p_lock);
744 744 sigtoproc(p, server_thread, SIGCANCEL);
745 745 mutex_exit(&p->p_lock);
746 746
747 747 mutex_enter(&door_knob);
748 748 DOOR_T_RELEASE(st);
749 749 }
750 750 }
751 751 } else {
752 752 /*
753 753 * Return from stop(), server exit...
754 754 *
755 755 * Note that the server could have done a
756 756 * door_return while the client was in stop state
757 757 * (ISSIG), in which case the error condition
758 758 * is updated by the server.
759 759 */
760 760 mutex_enter(&door_knob);
761 761 if (ct->d_error == DOOR_WAIT) {
762 762 /* Still waiting for a reply */
763 763 shuttle_swtch(&door_knob);
764 764 mutex_enter(&door_knob);
765 765 lwp->lwp_asleep = 0;
766 766 goto shuttle_return;
767 767 } else if (ct->d_error == DOOR_EXIT) {
768 768 /* Server exit */
769 769 error = EINTR;
770 770 } else {
771 771 /* Server did a door_return during ISSIG */
772 772 error = ct->d_error;
773 773 }
774 774 }
775 775 /*
776 776 * Can't exit if the server is currently copying
777 777 * results for me.
778 778 */
779 779 while (DOOR_T_HELD(ct))
780 780 cv_wait(&ct->d_cv, &door_knob);
781 781
782 782 /*
783 783 * If the server has not processed our message, free the
784 784 * descriptors.
785 785 */
786 786 if (!ct->d_args_done) {
787 787 needcleanup = 1;
788 788 ct->d_args_done = 1;
789 789 }
790 790
791 791 /*
792 792 * Find out if results were successfully copied.
793 793 */
794 794 if (ct->d_error == 0)
795 795 gotresults = 1;
796 796 }
797 797 ASSERT(ct->d_args_done);
798 798 lwp->lwp_asleep = 0; /* /proc */
799 799 lwp->lwp_sysabort = 0; /* /proc */
800 800 if (--dp->door_active == 0 && (dp->door_flags & DOOR_DELAY))
801 801 door_deliver_unref(dp);
802 802 mutex_exit(&door_knob);
803 803
804 804 if (needcleanup)
805 805 door_fp_close(ct->d_fpp, ct->d_args.desc_num);
806 806
807 807 results:
808 808 /*
809 809 * Move the results to userland (if any)
810 810 */
811 811
812 812 if (ct->d_noresults)
813 813 goto out;
814 814
815 815 if (error) {
816 816 /*
817 817 * If server returned results successfully, then we've
818 818 * been interrupted and may need to clean up.
819 819 */
820 820 if (gotresults) {
821 821 ASSERT(error == EINTR);
822 822 door_fp_close(ct->d_fpp, ct->d_args.desc_num);
823 823 }
824 824 goto out;
825 825 }
826 826
827 827 /*
828 828 * Copy back data if we haven't caused an overflow (already
829 829 * handled) and we are using a 2 copy transfer, or we are
830 830 * returning data from a kernel server.
831 831 */
832 832 if (ct->d_args.data_size) {
833 833 ct->d_args.data_ptr = ct->d_args.rbuf;
834 834 if (ct->d_kernel || (!ct->d_overflow &&
835 835 ct->d_args.data_size <= door_max_arg)) {
836 836 if (copyout_nowatch(ct->d_buf, ct->d_args.rbuf,
837 837 ct->d_args.data_size)) {
838 838 door_fp_close(ct->d_fpp, ct->d_args.desc_num);
839 839 error = EFAULT;
840 840 goto out;
841 841 }
842 842 }
843 843 }
844 844
845 845 /*
846 846 * stuff returned doors into our proc, copyout the descriptors
847 847 */
848 848 if (ct->d_args.desc_num) {
849 849 struct file **fpp;
850 850 door_desc_t *didpp;
851 851 uint_t n = ct->d_args.desc_num;
852 852
853 853 dsize = n * sizeof (door_desc_t);
854 854 start = didpp = kmem_alloc(dsize, KM_SLEEP);
855 855 fpp = ct->d_fpp;
856 856
857 857 while (n--) {
858 858 if (door_insert(*fpp, didpp) == -1) {
859 859 /* Close remaining files */
860 860 door_fp_close(fpp, n + 1);
861 861 error = EMFILE;
862 862 goto out;
863 863 }
864 864 fpp++; didpp++; ncopied++;
865 865 }
866 866
867 867 ct->d_args.desc_ptr = (door_desc_t *)(ct->d_args.rbuf +
868 868 roundup(ct->d_args.data_size, sizeof (door_desc_t)));
869 869
870 870 if (copyout_nowatch(start, ct->d_args.desc_ptr, dsize)) {
871 871 error = EFAULT;
872 872 goto out;
873 873 }
874 874 }
875 875
876 876 /*
877 877 * Return the results
878 878 */
879 879 if (datamodel == DATAMODEL_NATIVE) {
880 880 if (copyout_nowatch(&ct->d_args, args,
881 881 sizeof (door_arg_t)) != 0)
882 882 error = EFAULT;
883 883 } else {
884 884 door_arg32_t da32;
885 885
886 886 da32.data_ptr = (caddr32_t)(uintptr_t)ct->d_args.data_ptr;
887 887 da32.data_size = ct->d_args.data_size;
888 888 da32.desc_ptr = (caddr32_t)(uintptr_t)ct->d_args.desc_ptr;
889 889 da32.desc_num = ct->d_args.desc_num;
890 890 da32.rbuf = (caddr32_t)(uintptr_t)ct->d_args.rbuf;
891 891 da32.rsize = ct->d_args.rsize;
892 892 if (copyout_nowatch(&da32, args, sizeof (door_arg32_t)) != 0) {
893 893 error = EFAULT;
894 894 }
895 895 }
896 896
897 897 out:
898 898 ct->d_noresults = 0;
899 899
900 900 /* clean up the overflow buffer if an error occurred */
901 901 if (error != 0 && ct->d_overflow) {
902 902 (void) as_unmap(curproc->p_as, ct->d_args.rbuf,
903 903 ct->d_args.rsize);
904 904 }
905 905 ct->d_overflow = 0;
906 906
907 907 /* call destructor */
908 908 if (destfn) {
909 909 ASSERT(ct->d_kernel);
910 910 (*destfn)(dp->door_data, destarg);
911 911 ct->d_buf = NULL;
912 912 ct->d_bufsize = 0;
913 913 }
914 914
915 915 if (dp)
916 916 VN_RELE(DTOV(dp));
917 917
918 918 if (ct->d_buf) {
919 919 ASSERT(!ct->d_kernel);
920 920 kmem_free(ct->d_buf, ct->d_bufsize);
921 921 ct->d_buf = NULL;
922 922 ct->d_bufsize = 0;
923 923 }
924 924 ct->d_kernel = 0;
925 925
926 926 /* clean up the descriptor copyout buffer */
927 927 if (start != NULL) {
928 928 if (error != 0)
929 929 door_fd_close(start, ncopied);
930 930 kmem_free(start, dsize);
931 931 }
932 932
933 933 if (ct->d_fpp) {
934 934 kmem_free(ct->d_fpp, ct->d_fpp_size);
935 935 ct->d_fpp = NULL;
936 936 ct->d_fpp_size = 0;
937 937 }
938 938
939 939 if (error)
940 940 return (set_errno(error));
941 941
942 942 return (0);
943 943 }
944 944
945 945 static int
946 946 door_setparam_common(door_node_t *dp, int from_kernel, int type, size_t val)
947 947 {
948 948 int error = 0;
949 949
950 950 mutex_enter(&door_knob);
951 951
952 952 if (DOOR_INVALID(dp)) {
953 953 mutex_exit(&door_knob);
954 954 return (EBADF);
955 955 }
956 956
957 957 /*
958 958 * door_ki_setparam() can only affect kernel doors.
959 959 * door_setparam() can only affect doors attached to the current
960 960 * process.
961 961 */
962 962 if ((from_kernel && dp->door_target != &p0) ||
963 963 (!from_kernel && dp->door_target != curproc)) {
964 964 mutex_exit(&door_knob);
965 965 return (EPERM);
966 966 }
967 967
968 968 switch (type) {
969 969 case DOOR_PARAM_DESC_MAX:
970 970 if (val > INT_MAX)
971 971 error = ERANGE;
972 972 else if ((dp->door_flags & DOOR_REFUSE_DESC) && val != 0)
973 973 error = ENOTSUP;
974 974 else
975 975 dp->door_desc_max = (uint_t)val;
976 976 break;
977 977
978 978 case DOOR_PARAM_DATA_MIN:
979 979 if (val > dp->door_data_max)
980 980 error = EINVAL;
981 981 else
982 982 dp->door_data_min = val;
983 983 break;
984 984
985 985 case DOOR_PARAM_DATA_MAX:
986 986 if (val < dp->door_data_min)
987 987 error = EINVAL;
988 988 else
989 989 dp->door_data_max = val;
990 990 break;
991 991
992 992 default:
993 993 error = EINVAL;
994 994 break;
995 995 }
996 996
997 997 mutex_exit(&door_knob);
998 998 return (error);
999 999 }
1000 1000
1001 1001 static int
1002 1002 door_getparam_common(door_node_t *dp, int type, size_t *out)
1003 1003 {
1004 1004 int error = 0;
1005 1005
1006 1006 mutex_enter(&door_knob);
1007 1007 switch (type) {
1008 1008 case DOOR_PARAM_DESC_MAX:
1009 1009 *out = (size_t)dp->door_desc_max;
1010 1010 break;
1011 1011 case DOOR_PARAM_DATA_MIN:
1012 1012 *out = dp->door_data_min;
1013 1013 break;
1014 1014 case DOOR_PARAM_DATA_MAX:
1015 1015 *out = dp->door_data_max;
1016 1016 break;
1017 1017 default:
1018 1018 error = EINVAL;
1019 1019 break;
1020 1020 }
1021 1021 mutex_exit(&door_knob);
1022 1022 return (error);
1023 1023 }
1024 1024
1025 1025 int
1026 1026 door_setparam(int did, int type, size_t val)
1027 1027 {
1028 1028 door_node_t *dp;
1029 1029 int error = 0;
1030 1030
1031 1031 if ((dp = door_lookup(did, NULL)) == NULL)
1032 1032 return (set_errno(EBADF));
1033 1033
1034 1034 error = door_setparam_common(dp, 0, type, val);
1035 1035
1036 1036 releasef(did);
1037 1037
1038 1038 if (error)
1039 1039 return (set_errno(error));
1040 1040
1041 1041 return (0);
1042 1042 }
1043 1043
1044 1044 int
1045 1045 door_getparam(int did, int type, size_t *out)
1046 1046 {
1047 1047 door_node_t *dp;
1048 1048 size_t val = 0;
1049 1049 int error = 0;
1050 1050
1051 1051 if ((dp = door_lookup(did, NULL)) == NULL)
1052 1052 return (set_errno(EBADF));
1053 1053
1054 1054 error = door_getparam_common(dp, type, &val);
1055 1055
1056 1056 releasef(did);
1057 1057
1058 1058 if (error)
1059 1059 return (set_errno(error));
1060 1060
1061 1061 if (get_udatamodel() == DATAMODEL_NATIVE) {
1062 1062 if (copyout(&val, out, sizeof (val)))
1063 1063 return (set_errno(EFAULT));
1064 1064 #ifdef _SYSCALL32_IMPL
1065 1065 } else {
1066 1066 size32_t val32 = (size32_t)val;
1067 1067
1068 1068 if (val != val32)
1069 1069 return (set_errno(EOVERFLOW));
1070 1070
1071 1071 if (copyout(&val32, out, sizeof (val32)))
1072 1072 return (set_errno(EFAULT));
1073 1073 #endif /* _SYSCALL32_IMPL */
1074 1074 }
1075 1075
1076 1076 return (0);
1077 1077 }
1078 1078
1079 1079 /*
1080 1080 * A copyout() which proceeds from high addresses to low addresses. This way,
1081 1081 * stack guard pages are effective.
1082 1082 *
1083 1083 * Note that we use copyout_nowatch(); this is called while the client is
1084 1084 * held.
1085 1085 */
1086 1086 static int
1087 1087 door_stack_copyout(const void *kaddr, void *uaddr, size_t count)
1088 1088 {
1089 1089 const char *kbase = (const char *)kaddr;
1090 1090 uintptr_t ubase = (uintptr_t)uaddr;
1091 1091 size_t pgsize = PAGESIZE;
1092 1092
1093 1093 if (count <= pgsize)
1094 1094 return (copyout_nowatch(kaddr, uaddr, count));
1095 1095
1096 1096 while (count > 0) {
1097 1097 uintptr_t start, end, offset, amount;
1098 1098
1099 1099 end = ubase + count;
1100 1100 start = P2ALIGN(end - 1, pgsize);
1101 1101 if (P2ALIGN(ubase, pgsize) == start)
1102 1102 start = ubase;
1103 1103
1104 1104 offset = start - ubase;
1105 1105 amount = end - start;
1106 1106
1107 1107 ASSERT(amount > 0 && amount <= count && amount <= pgsize);
1108 1108
1109 1109 if (copyout_nowatch(kbase + offset, (void *)start, amount))
1110 1110 return (1);
1111 1111 count -= amount;
1112 1112 }
1113 1113 return (0);
1114 1114 }
1115 1115
1116 1116 /*
1117 1117 * Writes the stack layout for door_return() into the door_server_t of the
1118 1118 * server thread.
1119 1119 */
1120 1120 static int
1121 1121 door_layout(kthread_t *tp, size_t data_size, uint_t ndesc, int info_needed)
1122 1122 {
1123 1123 door_server_t *st = DOOR_SERVER(tp->t_door);
1124 1124 door_layout_t *out = &st->d_layout;
1125 1125 uintptr_t base_sp = (uintptr_t)st->d_sp;
1126 1126 size_t ssize = st->d_ssize;
1127 1127 size_t descsz;
1128 1128 uintptr_t descp, datap, infop, resultsp, finalsp;
1129 1129 size_t align = STACK_ALIGN;
1130 1130 size_t results_sz = sizeof (struct door_results);
1131 1131 model_t datamodel = lwp_getdatamodel(ttolwp(tp));
1132 1132
1133 1133 ASSERT(!st->d_layout_done);
1134 1134
1135 1135 #ifndef _STACK_GROWS_DOWNWARD
1136 1136 #error stack does not grow downward, door_layout() must change
1137 1137 #endif
1138 1138
1139 1139 #ifdef _SYSCALL32_IMPL
1140 1140 if (datamodel != DATAMODEL_NATIVE) {
1141 1141 align = STACK_ALIGN32;
1142 1142 results_sz = sizeof (struct door_results32);
1143 1143 }
1144 1144 #endif
1145 1145
1146 1146 descsz = ndesc * sizeof (door_desc_t);
1147 1147
1148 1148 /*
1149 1149 * To speed up the overflow checking, we do an initial check
1150 1150 * that the passed in data size won't cause us to wrap past
1151 1151 * base_sp. Since door_max_desc limits descsz, we can
1152 1152 * safely use it here. 65535 is an arbitrary 'bigger than
1153 1153 * we need, small enough to not cause trouble' constant;
1154 1154 * the only constraint is that it must be > than:
1155 1155 *
1156 1156 * 5 * STACK_ALIGN +
1157 1157 * sizeof (door_info_t) +
1158 1158 * sizeof (door_results_t) +
1159 1159 * (max adjustment from door_final_sp())
1160 1160 *
1161 1161 * After we compute the layout, we can safely do a "did we wrap
1162 1162 * around" check, followed by a check against the recorded
1163 1163 * stack size.
1164 1164 */
1165 1165 if (data_size >= SIZE_MAX - (size_t)65535UL - descsz)
1166 1166 return (E2BIG); /* overflow */
1167 1167
1168 1168 descp = P2ALIGN(base_sp - descsz, align);
1169 1169 datap = P2ALIGN(descp - data_size, align);
1170 1170
1171 1171 if (info_needed)
1172 1172 infop = P2ALIGN(datap - sizeof (door_info_t), align);
1173 1173 else
1174 1174 infop = datap;
1175 1175
1176 1176 resultsp = P2ALIGN(infop - results_sz, align);
1177 1177 finalsp = door_final_sp(resultsp, align, datamodel);
1178 1178
1179 1179 if (finalsp > base_sp)
1180 1180 return (E2BIG); /* overflow */
1181 1181
1182 1182 if (ssize != 0 && (base_sp - finalsp) > ssize)
1183 1183 return (E2BIG); /* doesn't fit in stack */
1184 1184
1185 1185 out->dl_descp = (ndesc != 0)? (caddr_t)descp : 0;
1186 1186 out->dl_datap = (data_size != 0)? (caddr_t)datap : 0;
1187 1187 out->dl_infop = info_needed? (caddr_t)infop : 0;
1188 1188 out->dl_resultsp = (caddr_t)resultsp;
1189 1189 out->dl_sp = (caddr_t)finalsp;
1190 1190
1191 1191 st->d_layout_done = 1;
1192 1192 return (0);
1193 1193 }
1194 1194
1195 1195 static int
1196 1196 door_server_dispatch(door_client_t *ct, door_node_t *dp)
1197 1197 {
1198 1198 door_server_t *st = DOOR_SERVER(curthread->t_door);
1199 1199 door_layout_t *layout = &st->d_layout;
1200 1200 int error = 0;
1201 1201
1202 1202 int is_private = (dp->door_flags & DOOR_PRIVATE);
1203 1203
1204 1204 door_pool_t *pool = (is_private)? &dp->door_servers :
1205 1205 &curproc->p_server_threads;
1206 1206
1207 1207 int empty_pool = (pool->dp_threads == NULL);
1208 1208
1209 1209 caddr_t infop = NULL;
1210 1210 char *datap = NULL;
1211 1211 size_t datasize = 0;
1212 1212 size_t descsize;
1213 1213
1214 1214 file_t **fpp = ct->d_fpp;
1215 1215 door_desc_t *start = NULL;
1216 1216 uint_t ndesc = 0;
1217 1217 uint_t ncopied = 0;
1218 1218
1219 1219 if (ct != NULL) {
1220 1220 datap = ct->d_args.data_ptr;
1221 1221 datasize = ct->d_args.data_size;
1222 1222 ndesc = ct->d_args.desc_num;
1223 1223 }
1224 1224
1225 1225 descsize = ndesc * sizeof (door_desc_t);
1226 1226
1227 1227 /*
1228 1228 * Reset datap to NULL if we aren't passing any data. Be careful
1229 1229 * to let unref notifications through, though.
1230 1230 */
1231 1231 if (datap == DOOR_UNREF_DATA) {
1232 1232 if (ct->d_upcall != NULL)
1233 1233 datasize = 0;
1234 1234 else
1235 1235 datap = NULL;
1236 1236 } else if (datasize == 0) {
1237 1237 datap = NULL;
1238 1238 }
1239 1239
1240 1240 /*
1241 1241 * Get the stack layout, if it hasn't already been done.
1242 1242 */
1243 1243 if (!st->d_layout_done) {
1244 1244 error = door_layout(curthread, datasize, ndesc,
1245 1245 (is_private && empty_pool));
1246 1246 if (error != 0)
1247 1247 goto fail;
1248 1248 }
1249 1249
1250 1250 /*
1251 1251 * fill out the stack, starting from the top. Layout was already
1252 1252 * filled in by door_args() or door_translate_out().
1253 1253 */
1254 1254 if (layout->dl_descp != NULL) {
1255 1255 ASSERT(ndesc != 0);
1256 1256 start = kmem_alloc(descsize, KM_SLEEP);
1257 1257
1258 1258 while (ndesc > 0) {
1259 1259 if (door_insert(*fpp, &start[ncopied]) == -1) {
1260 1260 error = EMFILE;
1261 1261 goto fail;
1262 1262 }
1263 1263 ndesc--;
1264 1264 ncopied++;
1265 1265 fpp++;
1266 1266 }
1267 1267 if (door_stack_copyout(start, layout->dl_descp, descsize)) {
1268 1268 error = E2BIG;
1269 1269 goto fail;
1270 1270 }
1271 1271 }
1272 1272 fpp = NULL; /* finished processing */
1273 1273
1274 1274 if (layout->dl_datap != NULL) {
1275 1275 ASSERT(datasize != 0);
1276 1276 datap = layout->dl_datap;
1277 1277 if (ct->d_upcall != NULL || datasize <= door_max_arg) {
1278 1278 if (door_stack_copyout(ct->d_buf, datap, datasize)) {
1279 1279 error = E2BIG;
1280 1280 goto fail;
1281 1281 }
1282 1282 }
1283 1283 }
1284 1284
1285 1285 if (is_private && empty_pool) {
1286 1286 door_info_t di;
1287 1287
1288 1288 infop = layout->dl_infop;
1289 1289 ASSERT(infop != NULL);
1290 1290
1291 1291 di.di_target = curproc->p_pid;
1292 1292 di.di_proc = (door_ptr_t)(uintptr_t)dp->door_pc;
1293 1293 di.di_data = (door_ptr_t)(uintptr_t)dp->door_data;
1294 1294 di.di_uniquifier = dp->door_index;
1295 1295 di.di_attributes = (dp->door_flags & DOOR_ATTR_MASK) |
1296 1296 DOOR_LOCAL;
1297 1297
1298 1298 if (door_stack_copyout(&di, infop, sizeof (di))) {
1299 1299 error = E2BIG;
1300 1300 goto fail;
1301 1301 }
1302 1302 }
1303 1303
1304 1304 if (get_udatamodel() == DATAMODEL_NATIVE) {
1305 1305 struct door_results dr;
1306 1306
1307 1307 dr.cookie = dp->door_data;
1308 1308 dr.data_ptr = datap;
1309 1309 dr.data_size = datasize;
1310 1310 dr.desc_ptr = (door_desc_t *)layout->dl_descp;
1311 1311 dr.desc_num = ncopied;
1312 1312 dr.pc = dp->door_pc;
1313 1313 dr.nservers = !empty_pool;
1314 1314 dr.door_info = (door_info_t *)infop;
1315 1315
1316 1316 if (door_stack_copyout(&dr, layout->dl_resultsp, sizeof (dr))) {
1317 1317 error = E2BIG;
1318 1318 goto fail;
1319 1319 }
1320 1320 #ifdef _SYSCALL32_IMPL
1321 1321 } else {
1322 1322 struct door_results32 dr32;
1323 1323
1324 1324 dr32.cookie = (caddr32_t)(uintptr_t)dp->door_data;
1325 1325 dr32.data_ptr = (caddr32_t)(uintptr_t)datap;
1326 1326 dr32.data_size = (size32_t)datasize;
1327 1327 dr32.desc_ptr = (caddr32_t)(uintptr_t)layout->dl_descp;
1328 1328 dr32.desc_num = ncopied;
1329 1329 dr32.pc = (caddr32_t)(uintptr_t)dp->door_pc;
1330 1330 dr32.nservers = !empty_pool;
1331 1331 dr32.door_info = (caddr32_t)(uintptr_t)infop;
1332 1332
1333 1333 if (door_stack_copyout(&dr32, layout->dl_resultsp,
1334 1334 sizeof (dr32))) {
1335 1335 error = E2BIG;
1336 1336 goto fail;
1337 1337 }
1338 1338 #endif
1339 1339 }
1340 1340
1341 1341 error = door_finish_dispatch(layout->dl_sp);
1342 1342 fail:
1343 1343 if (start != NULL) {
1344 1344 if (error != 0)
1345 1345 door_fd_close(start, ncopied);
1346 1346 kmem_free(start, descsize);
1347 1347 }
1348 1348 if (fpp != NULL)
1349 1349 door_fp_close(fpp, ndesc);
1350 1350
1351 1351 return (error);
1352 1352 }
1353 1353
1354 1354 /*
1355 1355 * Return the results (if any) to the caller (if any) and wait for the
1356 1356 * next invocation on a door.
1357 1357 */
1358 1358 int
1359 1359 door_return(caddr_t data_ptr, size_t data_size,
1360 1360 door_desc_t *desc_ptr, uint_t desc_num, caddr_t sp, size_t ssize)
1361 1361 {
1362 1362 kthread_t *caller;
1363 1363 klwp_t *lwp;
1364 1364 int error = 0;
1365 1365 door_node_t *dp;
1366 1366 door_server_t *st; /* curthread door_data */
1367 1367 door_client_t *ct; /* caller door_data */
1368 1368 int cancel_pending;
1369 1369
1370 1370 st = door_my_server(1);
1371 1371
1372 1372 /*
1373 1373 * If thread was bound to a door that no longer exists, return
1374 1374 * an error. This can happen if a thread is bound to a door
1375 1375 * before the process calls forkall(); in the child, the door
1376 1376 * doesn't exist and door_fork() sets the d_invbound flag.
1377 1377 */
1378 1378 if (st->d_invbound)
1379 1379 return (set_errno(EINVAL));
1380 1380
1381 1381 st->d_sp = sp; /* Save base of stack. */
1382 1382 st->d_ssize = ssize; /* and its size */
1383 1383
1384 1384 /*
1385 1385 * This should be done in shuttle_resume(), just before going to
1386 1386 * sleep, but we want to avoid overhead while holding door_knob.
1387 1387 * prstop() is just a no-op if we don't really go to sleep.
1388 1388 * We test not-kernel-address-space for the sake of clustering code.
1389 1389 */
1390 1390 lwp = ttolwp(curthread);
1391 1391 if (lwp && lwp->lwp_nostop == 0 && curproc->p_as != &kas)
1392 1392 prstop(PR_REQUESTED, 0);
1393 1393
1394 1394 /* Make sure the caller hasn't gone away */
1395 1395 mutex_enter(&door_knob);
1396 1396 if ((caller = st->d_caller) == NULL || caller->t_door == NULL) {
1397 1397 if (desc_num != 0) {
1398 1398 /* close any DOOR_RELEASE descriptors */
1399 1399 mutex_exit(&door_knob);
1400 1400 error = door_release_fds(desc_ptr, desc_num);
1401 1401 if (error)
1402 1402 return (set_errno(error));
1403 1403 mutex_enter(&door_knob);
1404 1404 }
1405 1405 goto out;
1406 1406 }
1407 1407 ct = DOOR_CLIENT(caller->t_door);
1408 1408
1409 1409 ct->d_args.data_size = data_size;
1410 1410 ct->d_args.desc_num = desc_num;
1411 1411 /*
1412 1412 * Transfer results, if any, to the client
1413 1413 */
1414 1414 if (data_size != 0 || desc_num != 0) {
1415 1415 /*
1416 1416 * Prevent the client from exiting until we have finished
1417 1417 * moving results.
1418 1418 */
1419 1419 DOOR_T_HOLD(ct);
1420 1420 mutex_exit(&door_knob);
1421 1421 error = door_results(caller, data_ptr, data_size,
1422 1422 desc_ptr, desc_num);
1423 1423 mutex_enter(&door_knob);
1424 1424 DOOR_T_RELEASE(ct);
1425 1425 /*
1426 1426 * Pass EOVERFLOW errors back to the client
1427 1427 */
1428 1428 if (error && error != EOVERFLOW) {
1429 1429 mutex_exit(&door_knob);
1430 1430 return (set_errno(error));
1431 1431 }
1432 1432 }
1433 1433 out:
1434 1434 /* Put ourselves on the available server thread list */
1435 1435 door_release_server(st->d_pool, curthread);
1436 1436
1437 1437 /*
1438 1438 * Make sure the caller is still waiting to be resumed
1439 1439 */
1440 1440 if (caller) {
1441 1441 disp_lock_t *tlp;
1442 1442
1443 1443 thread_lock(caller);
1444 1444 ct->d_error = error; /* Return any errors */
1445 1445 if (caller->t_state == TS_SLEEP &&
1446 1446 SOBJ_TYPE(caller->t_sobj_ops) == SOBJ_SHUTTLE) {
1447 1447 cpu_t *cp = CPU;
1448 1448
1449 1449 tlp = caller->t_lockp;
1450 1450 /*
1451 1451 * Setting t_disp_queue prevents erroneous preemptions
1452 1452 * if this thread is still in execution on another
1453 1453 * processor
1454 1454 */
1455 1455 caller->t_disp_queue = cp->cpu_disp;
1456 1456 CL_ACTIVE(caller);
1457 1457 /*
1458 1458 * We are calling thread_onproc() instead of
1459 1459 * THREAD_ONPROC() because compiler can reorder
1460 1460 * the two stores of t_state and t_lockp in
1461 1461 * THREAD_ONPROC().
1462 1462 */
1463 1463 thread_onproc(caller, cp);
1464 1464 disp_lock_exit_high(tlp);
1465 1465 shuttle_resume(caller, &door_knob);
1466 1466 } else {
1467 1467 /* May have been setrun or in stop state */
1468 1468 thread_unlock(caller);
1469 1469 shuttle_swtch(&door_knob);
1470 1470 }
1471 1471 } else {
1472 1472 shuttle_swtch(&door_knob);
1473 1473 }
1474 1474
1475 1475 /*
1476 1476 * We've sprung to life. Determine if we are part of a door
1477 1477 * invocation, or just interrupted
1478 1478 */
1479 1479 mutex_enter(&door_knob);
1480 1480 if ((dp = st->d_active) != NULL) {
1481 1481 /*
1482 1482 * Normal door invocation. Return any error condition
1483 1483 * encountered while trying to pass args to the server
1484 1484 * thread.
1485 1485 */
1486 1486 lwp->lwp_asleep = 0;
1487 1487 /*
1488 1488 * Prevent the caller from leaving us while we
1489 1489 * are copying out the arguments from it's buffer.
1490 1490 */
1491 1491 ASSERT(st->d_caller != NULL);
1492 1492 ct = DOOR_CLIENT(st->d_caller->t_door);
1493 1493
1494 1494 DOOR_T_HOLD(ct);
1495 1495 mutex_exit(&door_knob);
1496 1496 error = door_server_dispatch(ct, dp);
1497 1497 mutex_enter(&door_knob);
1498 1498 DOOR_T_RELEASE(ct);
1499 1499
1500 1500 /* let the client know we have processed its message */
1501 1501 ct->d_args_done = 1;
1502 1502
1503 1503 if (error) {
1504 1504 caller = st->d_caller;
1505 1505 if (caller)
1506 1506 ct = DOOR_CLIENT(caller->t_door);
1507 1507 else
1508 1508 ct = NULL;
1509 1509 goto out;
1510 1510 }
1511 1511 mutex_exit(&door_knob);
1512 1512 return (0);
1513 1513 } else {
1514 1514 /*
1515 1515 * We are not involved in a door_invocation.
1516 1516 * Check for /proc related activity...
1517 1517 */
1518 1518 st->d_caller = NULL;
1519 1519 door_server_exit(curproc, curthread);
1520 1520 mutex_exit(&door_knob);
1521 1521 cancel_pending = 0;
1522 1522 if (ISSIG(curthread, FORREAL) || lwp->lwp_sysabort ||
1523 1523 MUSTRETURN(curproc, curthread) ||
1524 1524 (cancel_pending = schedctl_cancel_pending()) != 0) {
1525 1525 if (cancel_pending)
1526 1526 schedctl_cancel_eintr();
1527 1527 lwp->lwp_asleep = 0;
1528 1528 lwp->lwp_sysabort = 0;
1529 1529 return (set_errno(EINTR));
1530 1530 }
1531 1531 /* Go back and wait for another request */
1532 1532 lwp->lwp_asleep = 0;
1533 1533 mutex_enter(&door_knob);
1534 1534 caller = NULL;
1535 1535 goto out;
1536 1536 }
1537 1537 }
1538 1538
1539 1539 /*
1540 1540 * Revoke any future invocations on this door
1541 1541 */
1542 1542 int
1543 1543 door_revoke(int did)
1544 1544 {
1545 1545 door_node_t *d;
1546 1546 int error;
1547 1547
1548 1548 if ((d = door_lookup(did, NULL)) == NULL)
1549 1549 return (set_errno(EBADF));
1550 1550
1551 1551 mutex_enter(&door_knob);
1552 1552 if (d->door_target != curproc) {
1553 1553 mutex_exit(&door_knob);
1554 1554 releasef(did);
1555 1555 return (set_errno(EPERM));
1556 1556 }
1557 1557 d->door_flags |= DOOR_REVOKED;
1558 1558 if (d->door_flags & DOOR_PRIVATE)
1559 1559 cv_broadcast(&d->door_servers.dp_cv);
1560 1560 else
1561 1561 cv_broadcast(&curproc->p_server_threads.dp_cv);
1562 1562 mutex_exit(&door_knob);
1563 1563 releasef(did);
1564 1564 /* Invalidate the descriptor */
1565 1565 if ((error = closeandsetf(did, NULL)) != 0)
1566 1566 return (set_errno(error));
1567 1567 return (0);
1568 1568 }
1569 1569
1570 1570 int
1571 1571 door_info(int did, struct door_info *d_info)
1572 1572 {
1573 1573 door_node_t *dp;
1574 1574 door_info_t di;
1575 1575 door_server_t *st;
1576 1576 file_t *fp = NULL;
1577 1577
1578 1578 if (did == DOOR_QUERY) {
1579 1579 /* Get information on door current thread is bound to */
1580 1580 if ((st = door_my_server(0)) == NULL ||
1581 1581 (dp = st->d_pool) == NULL)
1582 1582 /* Thread isn't bound to a door */
1583 1583 return (set_errno(EBADF));
1584 1584 } else if ((dp = door_lookup(did, &fp)) == NULL) {
1585 1585 /* Not a door */
1586 1586 return (set_errno(EBADF));
1587 1587 }
1588 1588
1589 1589 door_info_common(dp, &di, fp);
1590 1590
1591 1591 if (did != DOOR_QUERY)
1592 1592 releasef(did);
1593 1593
1594 1594 if (copyout(&di, d_info, sizeof (struct door_info)))
1595 1595 return (set_errno(EFAULT));
1596 1596 return (0);
1597 1597 }
1598 1598
1599 1599 /*
1600 1600 * Common code for getting information about a door either via the
1601 1601 * door_info system call or the door_ki_info kernel call.
1602 1602 */
1603 1603 void
1604 1604 door_info_common(door_node_t *dp, struct door_info *dip, file_t *fp)
1605 1605 {
1606 1606 int unref_count;
1607 1607
1608 1608 bzero(dip, sizeof (door_info_t));
1609 1609
1610 1610 mutex_enter(&door_knob);
1611 1611 if (dp->door_target == NULL)
1612 1612 dip->di_target = -1;
1613 1613 else
1614 1614 dip->di_target = dp->door_target->p_pid;
1615 1615
1616 1616 dip->di_attributes = dp->door_flags & DOOR_ATTR_MASK;
1617 1617 if (dp->door_target == curproc)
1618 1618 dip->di_attributes |= DOOR_LOCAL;
1619 1619 dip->di_proc = (door_ptr_t)(uintptr_t)dp->door_pc;
1620 1620 dip->di_data = (door_ptr_t)(uintptr_t)dp->door_data;
1621 1621 dip->di_uniquifier = dp->door_index;
1622 1622 /*
1623 1623 * If this door is in the middle of having an unreferenced
1624 1624 * notification delivered, don't count the VN_HOLD by
1625 1625 * door_deliver_unref in determining if it is unreferenced.
1626 1626 * This handles the case where door_info is called from the
1627 1627 * thread delivering the unref notification.
1628 1628 */
1629 1629 if (dp->door_flags & DOOR_UNREF_ACTIVE)
1630 1630 unref_count = 2;
1631 1631 else
1632 1632 unref_count = 1;
1633 1633 mutex_exit(&door_knob);
1634 1634
1635 1635 if (fp == NULL) {
1636 1636 /*
1637 1637 * If this thread is bound to the door, then we can just
1638 1638 * check the vnode; a ref count of 1 (or 2 if this is
1639 1639 * handling an unref notification) means that the hold
1640 1640 * from the door_bind is the only reference to the door
1641 1641 * (no file descriptor refers to it).
1642 1642 */
1643 1643 if (DTOV(dp)->v_count == unref_count)
1644 1644 dip->di_attributes |= DOOR_IS_UNREF;
1645 1645 } else {
1646 1646 /*
1647 1647 * If we're working from a file descriptor or door handle
1648 1648 * we need to look at the file structure count. We don't
1649 1649 * need to hold the vnode lock since this is just a snapshot.
1650 1650 */
1651 1651 mutex_enter(&fp->f_tlock);
1652 1652 if (fp->f_count == 1 && DTOV(dp)->v_count == unref_count)
1653 1653 dip->di_attributes |= DOOR_IS_UNREF;
1654 1654 mutex_exit(&fp->f_tlock);
1655 1655 }
1656 1656 }
1657 1657
1658 1658 /*
1659 1659 * Return credentials of the door caller (if any) for this invocation
1660 1660 */
1661 1661 int
1662 1662 door_ucred(struct ucred_s *uch)
1663 1663 {
1664 1664 kthread_t *caller;
1665 1665 door_server_t *st;
1666 1666 door_client_t *ct;
1667 1667 door_upcall_t *dup;
1668 1668 struct proc *p;
1669 1669 struct ucred_s *res;
1670 1670 int err;
1671 1671
1672 1672 mutex_enter(&door_knob);
1673 1673 if ((st = door_my_server(0)) == NULL ||
1674 1674 (caller = st->d_caller) == NULL) {
1675 1675 mutex_exit(&door_knob);
1676 1676 return (set_errno(EINVAL));
1677 1677 }
1678 1678
1679 1679 ASSERT(caller->t_door != NULL);
1680 1680 ct = DOOR_CLIENT(caller->t_door);
1681 1681
1682 1682 /* Prevent caller from exiting while we examine the cred */
1683 1683 DOOR_T_HOLD(ct);
1684 1684 mutex_exit(&door_knob);
1685 1685
1686 1686 p = ttoproc(caller);
1687 1687
1688 1688 /*
1689 1689 * If the credentials are not specified by the client, get the one
1690 1690 * associated with the calling process.
1691 1691 */
1692 1692 if ((dup = ct->d_upcall) != NULL)
1693 1693 res = cred2ucred(dup->du_cred, p0.p_pid, NULL, CRED());
1694 1694 else
1695 1695 res = cred2ucred(caller->t_cred, p->p_pid, NULL, CRED());
1696 1696
1697 1697 mutex_enter(&door_knob);
1698 1698 DOOR_T_RELEASE(ct);
1699 1699 mutex_exit(&door_knob);
1700 1700
1701 1701 err = copyout(res, uch, res->uc_size);
1702 1702
1703 1703 kmem_free(res, res->uc_size);
1704 1704
1705 1705 if (err != 0)
1706 1706 return (set_errno(EFAULT));
1707 1707
1708 1708 return (0);
1709 1709 }
1710 1710
1711 1711 /*
1712 1712 * Bind the current lwp to the server thread pool associated with 'did'
1713 1713 */
1714 1714 int
1715 1715 door_bind(int did)
1716 1716 {
1717 1717 door_node_t *dp;
1718 1718 door_server_t *st;
1719 1719
1720 1720 if ((dp = door_lookup(did, NULL)) == NULL) {
1721 1721 /* Not a door */
1722 1722 return (set_errno(EBADF));
1723 1723 }
1724 1724
1725 1725 /*
1726 1726 * Can't bind to a non-private door, and can't bind to a door
1727 1727 * served by another process.
1728 1728 */
1729 1729 if ((dp->door_flags & DOOR_PRIVATE) == 0 ||
1730 1730 dp->door_target != curproc) {
1731 1731 releasef(did);
1732 1732 return (set_errno(EINVAL));
1733 1733 }
1734 1734
1735 1735 st = door_my_server(1);
1736 1736 if (st->d_pool)
1737 1737 door_unbind_thread(st->d_pool);
1738 1738 st->d_pool = dp;
1739 1739 st->d_invbound = 0;
1740 1740 door_bind_thread(dp);
1741 1741 releasef(did);
1742 1742
1743 1743 return (0);
1744 1744 }
1745 1745
1746 1746 /*
1747 1747 * Unbind the current lwp from it's server thread pool
1748 1748 */
1749 1749 int
1750 1750 door_unbind(void)
1751 1751 {
1752 1752 door_server_t *st;
1753 1753
1754 1754 if ((st = door_my_server(0)) == NULL)
1755 1755 return (set_errno(EBADF));
1756 1756
1757 1757 if (st->d_invbound) {
1758 1758 ASSERT(st->d_pool == NULL);
1759 1759 st->d_invbound = 0;
1760 1760 return (0);
1761 1761 }
1762 1762 if (st->d_pool == NULL)
1763 1763 return (set_errno(EBADF));
1764 1764 door_unbind_thread(st->d_pool);
1765 1765 st->d_pool = NULL;
1766 1766 return (0);
1767 1767 }
1768 1768
1769 1769 /*
1770 1770 * Create a descriptor for the associated file and fill in the
1771 1771 * attributes associated with it.
1772 1772 *
1773 1773 * Return 0 for success, -1 otherwise;
1774 1774 */
1775 1775 int
1776 1776 door_insert(struct file *fp, door_desc_t *dp)
1777 1777 {
1778 1778 struct vnode *vp;
1779 1779 int fd;
1780 1780 door_attr_t attributes = DOOR_DESCRIPTOR;
1781 1781
1782 1782 ASSERT(MUTEX_NOT_HELD(&door_knob));
1783 1783 if ((fd = ufalloc(0)) == -1)
1784 1784 return (-1);
1785 1785 setf(fd, fp);
1786 1786 dp->d_data.d_desc.d_descriptor = fd;
1787 1787
1788 1788 /* Fill in the attributes */
1789 1789 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
1790 1790 vp = fp->f_vnode;
1791 1791 if (vp && vp->v_type == VDOOR) {
1792 1792 if (VTOD(vp)->door_target == curproc)
1793 1793 attributes |= DOOR_LOCAL;
1794 1794 attributes |= VTOD(vp)->door_flags & DOOR_ATTR_MASK;
1795 1795 dp->d_data.d_desc.d_id = VTOD(vp)->door_index;
1796 1796 }
1797 1797 dp->d_attributes = attributes;
1798 1798 return (0);
1799 1799 }
1800 1800
1801 1801 /*
1802 1802 * Return an available thread for this server. A NULL return value indicates
1803 1803 * that either:
1804 1804 * The door has been revoked, or
1805 1805 * a signal was received.
1806 1806 * The two conditions can be differentiated using DOOR_INVALID(dp).
1807 1807 */
1808 1808 static kthread_t *
1809 1809 door_get_server(door_node_t *dp)
1810 1810 {
1811 1811 kthread_t **ktp;
1812 1812 kthread_t *server_t;
1813 1813 door_pool_t *pool;
1814 1814 door_server_t *st;
1815 1815 int signalled;
1816 1816
1817 1817 disp_lock_t *tlp;
1818 1818 cpu_t *cp;
1819 1819
1820 1820 ASSERT(MUTEX_HELD(&door_knob));
1821 1821
1822 1822 if (dp->door_flags & DOOR_PRIVATE)
1823 1823 pool = &dp->door_servers;
1824 1824 else
1825 1825 pool = &dp->door_target->p_server_threads;
1826 1826
1827 1827 for (;;) {
1828 1828 /*
1829 1829 * We search the thread pool, looking for a server thread
1830 1830 * ready to take an invocation (i.e. one which is still
1831 1831 * sleeping on a shuttle object). If none are available,
1832 1832 * we sleep on the pool's CV, and will be signaled when a
1833 1833 * thread is added to the pool.
1834 1834 *
1835 1835 * This relies on the fact that once a thread in the thread
1836 1836 * pool wakes up, it *must* remove and add itself to the pool
1837 1837 * before it can receive door calls.
1838 1838 */
1839 1839 if (DOOR_INVALID(dp))
1840 1840 return (NULL); /* Target has become invalid */
1841 1841
1842 1842 for (ktp = &pool->dp_threads;
1843 1843 (server_t = *ktp) != NULL;
1844 1844 ktp = &st->d_servers) {
1845 1845 st = DOOR_SERVER(server_t->t_door);
1846 1846
1847 1847 thread_lock(server_t);
1848 1848 if (server_t->t_state == TS_SLEEP &&
1849 1849 SOBJ_TYPE(server_t->t_sobj_ops) == SOBJ_SHUTTLE)
1850 1850 break;
1851 1851 thread_unlock(server_t);
1852 1852 }
1853 1853 if (server_t != NULL)
1854 1854 break; /* we've got a live one! */
1855 1855
1856 1856 if (!cv_wait_sig_swap_core(&pool->dp_cv, &door_knob,
1857 1857 &signalled)) {
1858 1858 /*
1859 1859 * If we were signaled and the door is still
1860 1860 * valid, pass the signal on to another waiter.
1861 1861 */
1862 1862 if (signalled && !DOOR_INVALID(dp))
1863 1863 cv_signal(&pool->dp_cv);
1864 1864 return (NULL); /* Got a signal */
1865 1865 }
1866 1866 }
1867 1867
1868 1868 /*
1869 1869 * We've got a thread_lock()ed thread which is still on the
1870 1870 * shuttle. Take it off the list of available server threads
1871 1871 * and mark it as ONPROC. We are committed to resuming this
1872 1872 * thread now.
1873 1873 */
1874 1874 tlp = server_t->t_lockp;
1875 1875 cp = CPU;
1876 1876
1877 1877 *ktp = st->d_servers;
1878 1878 st->d_servers = NULL;
1879 1879 /*
1880 1880 * Setting t_disp_queue prevents erroneous preemptions
1881 1881 * if this thread is still in execution on another processor
1882 1882 */
1883 1883 server_t->t_disp_queue = cp->cpu_disp;
1884 1884 CL_ACTIVE(server_t);
1885 1885 /*
1886 1886 * We are calling thread_onproc() instead of
1887 1887 * THREAD_ONPROC() because compiler can reorder
1888 1888 * the two stores of t_state and t_lockp in
1889 1889 * THREAD_ONPROC().
1890 1890 */
1891 1891 thread_onproc(server_t, cp);
1892 1892 disp_lock_exit(tlp);
1893 1893 return (server_t);
1894 1894 }
1895 1895
1896 1896 /*
1897 1897 * Put a server thread back in the pool.
1898 1898 */
1899 1899 static void
1900 1900 door_release_server(door_node_t *dp, kthread_t *t)
1901 1901 {
1902 1902 door_server_t *st = DOOR_SERVER(t->t_door);
1903 1903 door_pool_t *pool;
1904 1904
1905 1905 ASSERT(MUTEX_HELD(&door_knob));
1906 1906 st->d_active = NULL;
1907 1907 st->d_caller = NULL;
1908 1908 st->d_layout_done = 0;
1909 1909 if (dp && (dp->door_flags & DOOR_PRIVATE)) {
1910 1910 ASSERT(dp->door_target == NULL ||
1911 1911 dp->door_target == ttoproc(t));
1912 1912 pool = &dp->door_servers;
1913 1913 } else {
1914 1914 pool = &ttoproc(t)->p_server_threads;
1915 1915 }
1916 1916
1917 1917 st->d_servers = pool->dp_threads;
1918 1918 pool->dp_threads = t;
1919 1919
1920 1920 /* If someone is waiting for a server thread, wake him up */
1921 1921 cv_signal(&pool->dp_cv);
1922 1922 }
1923 1923
1924 1924 /*
1925 1925 * Remove a server thread from the pool if present.
1926 1926 */
1927 1927 static void
1928 1928 door_server_exit(proc_t *p, kthread_t *t)
1929 1929 {
1930 1930 door_pool_t *pool;
1931 1931 kthread_t **next;
1932 1932 door_server_t *st = DOOR_SERVER(t->t_door);
1933 1933
1934 1934 ASSERT(MUTEX_HELD(&door_knob));
1935 1935 if (st->d_pool != NULL) {
1936 1936 ASSERT(st->d_pool->door_flags & DOOR_PRIVATE);
1937 1937 pool = &st->d_pool->door_servers;
1938 1938 } else {
1939 1939 pool = &p->p_server_threads;
1940 1940 }
1941 1941
1942 1942 next = &pool->dp_threads;
1943 1943 while (*next != NULL) {
1944 1944 if (*next == t) {
1945 1945 *next = DOOR_SERVER(t->t_door)->d_servers;
1946 1946 return;
1947 1947 }
1948 1948 next = &(DOOR_SERVER((*next)->t_door)->d_servers);
1949 1949 }
1950 1950 }
1951 1951
1952 1952 /*
1953 1953 * Lookup the door descriptor. Caller must call releasef when finished
1954 1954 * with associated door.
1955 1955 */
1956 1956 static door_node_t *
1957 1957 door_lookup(int did, file_t **fpp)
1958 1958 {
1959 1959 vnode_t *vp;
1960 1960 file_t *fp;
1961 1961
1962 1962 ASSERT(MUTEX_NOT_HELD(&door_knob));
1963 1963 if ((fp = getf(did)) == NULL)
1964 1964 return (NULL);
1965 1965 /*
1966 1966 * Use the underlying vnode (we may be namefs mounted)
1967 1967 */
1968 1968 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
1969 1969 vp = fp->f_vnode;
1970 1970
1971 1971 if (vp == NULL || vp->v_type != VDOOR) {
1972 1972 releasef(did);
1973 1973 return (NULL);
1974 1974 }
1975 1975
1976 1976 if (fpp)
1977 1977 *fpp = fp;
1978 1978
1979 1979 return (VTOD(vp));
1980 1980 }
1981 1981
1982 1982 /*
1983 1983 * The current thread is exiting, so clean up any pending
1984 1984 * invocation details
1985 1985 */
1986 1986 void
1987 1987 door_slam(void)
1988 1988 {
1989 1989 door_node_t *dp;
1990 1990 door_data_t *dt;
1991 1991 door_client_t *ct;
1992 1992 door_server_t *st;
1993 1993
1994 1994 /*
1995 1995 * If we are an active door server, notify our
1996 1996 * client that we are exiting and revoke our door.
1997 1997 */
1998 1998 if ((dt = door_my_data(0)) == NULL)
1999 1999 return;
2000 2000 ct = DOOR_CLIENT(dt);
2001 2001 st = DOOR_SERVER(dt);
2002 2002
2003 2003 mutex_enter(&door_knob);
2004 2004 for (;;) {
2005 2005 if (DOOR_T_HELD(ct))
2006 2006 cv_wait(&ct->d_cv, &door_knob);
2007 2007 else if (DOOR_T_HELD(st))
2008 2008 cv_wait(&st->d_cv, &door_knob);
2009 2009 else
2010 2010 break; /* neither flag is set */
2011 2011 }
2012 2012 curthread->t_door = NULL;
2013 2013 if ((dp = st->d_active) != NULL) {
2014 2014 kthread_t *t = st->d_caller;
2015 2015 proc_t *p = curproc;
2016 2016
2017 2017 /* Revoke our door if the process is exiting */
2018 2018 if (dp->door_target == p && (p->p_flag & SEXITING)) {
2019 2019 door_list_delete(dp);
2020 2020 dp->door_target = NULL;
2021 2021 dp->door_flags |= DOOR_REVOKED;
2022 2022 if (dp->door_flags & DOOR_PRIVATE)
2023 2023 cv_broadcast(&dp->door_servers.dp_cv);
2024 2024 else
2025 2025 cv_broadcast(&p->p_server_threads.dp_cv);
2026 2026 }
2027 2027
2028 2028 if (t != NULL) {
2029 2029 /*
2030 2030 * Let the caller know we are gone
2031 2031 */
2032 2032 DOOR_CLIENT(t->t_door)->d_error = DOOR_EXIT;
2033 2033 thread_lock(t);
2034 2034 if (t->t_state == TS_SLEEP &&
2035 2035 SOBJ_TYPE(t->t_sobj_ops) == SOBJ_SHUTTLE)
2036 2036 setrun_locked(t);
2037 2037 thread_unlock(t);
2038 2038 }
2039 2039 }
2040 2040 mutex_exit(&door_knob);
2041 2041 if (st->d_pool)
2042 2042 door_unbind_thread(st->d_pool); /* Implicit door_unbind */
2043 2043 kmem_free(dt, sizeof (door_data_t));
2044 2044 }
2045 2045
2046 2046 /*
2047 2047 * Set DOOR_REVOKED for all doors of the current process. This is called
2048 2048 * on exit before all lwp's are being terminated so that door calls will
2049 2049 * return with an error.
2050 2050 */
2051 2051 void
2052 2052 door_revoke_all()
2053 2053 {
2054 2054 door_node_t *dp;
2055 2055 proc_t *p = ttoproc(curthread);
2056 2056
2057 2057 mutex_enter(&door_knob);
2058 2058 for (dp = p->p_door_list; dp != NULL; dp = dp->door_list) {
2059 2059 ASSERT(dp->door_target == p);
2060 2060 dp->door_flags |= DOOR_REVOKED;
2061 2061 if (dp->door_flags & DOOR_PRIVATE)
2062 2062 cv_broadcast(&dp->door_servers.dp_cv);
2063 2063 }
2064 2064 cv_broadcast(&p->p_server_threads.dp_cv);
2065 2065 mutex_exit(&door_knob);
2066 2066 }
2067 2067
2068 2068 /*
2069 2069 * The process is exiting, and all doors it created need to be revoked.
2070 2070 */
2071 2071 void
2072 2072 door_exit(void)
2073 2073 {
2074 2074 door_node_t *dp;
2075 2075 proc_t *p = ttoproc(curthread);
2076 2076
2077 2077 ASSERT(p->p_lwpcnt == 1);
2078 2078 /*
2079 2079 * Walk the list of active doors created by this process and
2080 2080 * revoke them all.
2081 2081 */
2082 2082 mutex_enter(&door_knob);
2083 2083 for (dp = p->p_door_list; dp != NULL; dp = dp->door_list) {
2084 2084 dp->door_target = NULL;
2085 2085 dp->door_flags |= DOOR_REVOKED;
2086 2086 if (dp->door_flags & DOOR_PRIVATE)
2087 2087 cv_broadcast(&dp->door_servers.dp_cv);
2088 2088 }
2089 2089 cv_broadcast(&p->p_server_threads.dp_cv);
2090 2090 /* Clear the list */
2091 2091 p->p_door_list = NULL;
2092 2092
2093 2093 /* Clean up the unref list */
2094 2094 while ((dp = p->p_unref_list) != NULL) {
2095 2095 p->p_unref_list = dp->door_ulist;
2096 2096 dp->door_ulist = NULL;
2097 2097 mutex_exit(&door_knob);
2098 2098 VN_RELE(DTOV(dp));
2099 2099 mutex_enter(&door_knob);
2100 2100 }
2101 2101 mutex_exit(&door_knob);
2102 2102 }
2103 2103
2104 2104
2105 2105 /*
2106 2106 * The process is executing forkall(), and we need to flag threads that
2107 2107 * are bound to a door in the child. This will make the child threads
2108 2108 * return an error to door_return unless they call door_unbind first.
2109 2109 */
2110 2110 void
2111 2111 door_fork(kthread_t *parent, kthread_t *child)
2112 2112 {
2113 2113 door_data_t *pt = parent->t_door;
2114 2114 door_server_t *st = DOOR_SERVER(pt);
2115 2115 door_data_t *dt;
2116 2116
2117 2117 ASSERT(MUTEX_NOT_HELD(&door_knob));
2118 2118 if (pt != NULL && (st->d_pool != NULL || st->d_invbound)) {
2119 2119 /* parent thread is bound to a door */
2120 2120 dt = child->t_door =
2121 2121 kmem_zalloc(sizeof (door_data_t), KM_SLEEP);
2122 2122 DOOR_SERVER(dt)->d_invbound = 1;
2123 2123 }
2124 2124 }
2125 2125
2126 2126 /*
2127 2127 * Deliver queued unrefs to appropriate door server.
2128 2128 */
2129 2129 static int
2130 2130 door_unref(void)
2131 2131 {
2132 2132 door_node_t *dp;
2133 2133 static door_arg_t unref_args = { DOOR_UNREF_DATA, 0, 0, 0, 0, 0 };
2134 2134 proc_t *p = ttoproc(curthread);
2135 2135
2136 2136 /* make sure there's only one unref thread per process */
2137 2137 mutex_enter(&door_knob);
2138 2138 if (p->p_unref_thread) {
2139 2139 mutex_exit(&door_knob);
2140 2140 return (set_errno(EALREADY));
2141 2141 }
2142 2142 p->p_unref_thread = 1;
2143 2143 mutex_exit(&door_knob);
2144 2144
2145 2145 (void) door_my_data(1); /* create info, if necessary */
2146 2146
2147 2147 for (;;) {
2148 2148 mutex_enter(&door_knob);
2149 2149
2150 2150 /* Grab a queued request */
2151 2151 while ((dp = p->p_unref_list) == NULL) {
2152 2152 if (!cv_wait_sig(&p->p_unref_cv, &door_knob)) {
2153 2153 /*
2154 2154 * Interrupted.
2155 2155 * Return so we can finish forkall() or exit().
2156 2156 */
2157 2157 p->p_unref_thread = 0;
2158 2158 mutex_exit(&door_knob);
2159 2159 return (set_errno(EINTR));
2160 2160 }
2161 2161 }
2162 2162 p->p_unref_list = dp->door_ulist;
2163 2163 dp->door_ulist = NULL;
2164 2164 dp->door_flags |= DOOR_UNREF_ACTIVE;
2165 2165 mutex_exit(&door_knob);
2166 2166
2167 2167 (void) door_upcall(DTOV(dp), &unref_args, NULL, SIZE_MAX, 0);
2168 2168
2169 2169 if (unref_args.rbuf != 0) {
2170 2170 kmem_free(unref_args.rbuf, unref_args.rsize);
2171 2171 unref_args.rbuf = NULL;
2172 2172 unref_args.rsize = 0;
2173 2173 }
2174 2174
2175 2175 mutex_enter(&door_knob);
2176 2176 ASSERT(dp->door_flags & DOOR_UNREF_ACTIVE);
2177 2177 dp->door_flags &= ~DOOR_UNREF_ACTIVE;
2178 2178 mutex_exit(&door_knob);
2179 2179 VN_RELE(DTOV(dp));
2180 2180 }
2181 2181 }
2182 2182
2183 2183
2184 2184 /*
2185 2185 * Deliver queued unrefs to kernel door server.
2186 2186 */
2187 2187 /* ARGSUSED */
2188 2188 static void
2189 2189 door_unref_kernel(caddr_t arg)
2190 2190 {
2191 2191 door_node_t *dp;
2192 2192 static door_arg_t unref_args = { DOOR_UNREF_DATA, 0, 0, 0, 0, 0 };
2193 2193 proc_t *p = ttoproc(curthread);
2194 2194 callb_cpr_t cprinfo;
2195 2195
2196 2196 /* should only be one of these */
2197 2197 mutex_enter(&door_knob);
2198 2198 if (p->p_unref_thread) {
2199 2199 mutex_exit(&door_knob);
2200 2200 return;
2201 2201 }
2202 2202 p->p_unref_thread = 1;
2203 2203 mutex_exit(&door_knob);
2204 2204
2205 2205 (void) door_my_data(1); /* make sure we have a door_data_t */
2206 2206
2207 2207 CALLB_CPR_INIT(&cprinfo, &door_knob, callb_generic_cpr, "door_unref");
2208 2208 for (;;) {
2209 2209 mutex_enter(&door_knob);
2210 2210 /* Grab a queued request */
2211 2211 while ((dp = p->p_unref_list) == NULL) {
2212 2212 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2213 2213 cv_wait(&p->p_unref_cv, &door_knob);
2214 2214 CALLB_CPR_SAFE_END(&cprinfo, &door_knob);
2215 2215 }
2216 2216 p->p_unref_list = dp->door_ulist;
2217 2217 dp->door_ulist = NULL;
2218 2218 dp->door_flags |= DOOR_UNREF_ACTIVE;
2219 2219 mutex_exit(&door_knob);
2220 2220
2221 2221 (*(dp->door_pc))(dp->door_data, &unref_args, NULL, NULL, NULL);
2222 2222
2223 2223 mutex_enter(&door_knob);
2224 2224 ASSERT(dp->door_flags & DOOR_UNREF_ACTIVE);
2225 2225 dp->door_flags &= ~DOOR_UNREF_ACTIVE;
2226 2226 mutex_exit(&door_knob);
2227 2227 VN_RELE(DTOV(dp));
2228 2228 }
2229 2229 }
2230 2230
2231 2231
2232 2232 /*
2233 2233 * Queue an unref invocation for processing for the current process
2234 2234 * The door may or may not be revoked at this point.
2235 2235 */
2236 2236 void
2237 2237 door_deliver_unref(door_node_t *d)
2238 2238 {
2239 2239 struct proc *server = d->door_target;
2240 2240
2241 2241 ASSERT(MUTEX_HELD(&door_knob));
2242 2242 ASSERT(d->door_active == 0);
2243 2243
2244 2244 if (server == NULL)
2245 2245 return;
2246 2246 /*
2247 2247 * Create a lwp to deliver unref calls if one isn't already running.
2248 2248 *
2249 2249 * A separate thread is used to deliver unrefs since the current
2250 2250 * thread may be holding resources (e.g. locks) in user land that
2251 2251 * may be needed by the unref processing. This would cause a
2252 2252 * deadlock.
2253 2253 */
2254 2254 if (d->door_flags & DOOR_UNREF_MULTI) {
2255 2255 /* multiple unrefs */
2256 2256 d->door_flags &= ~DOOR_DELAY;
2257 2257 } else {
2258 2258 /* Only 1 unref per door */
2259 2259 d->door_flags &= ~(DOOR_UNREF|DOOR_DELAY);
2260 2260 }
2261 2261 mutex_exit(&door_knob);
2262 2262
2263 2263 /*
2264 2264 * Need to bump the vnode count before putting the door on the
2265 2265 * list so it doesn't get prematurely released by door_unref.
2266 2266 */
2267 2267 VN_HOLD(DTOV(d));
2268 2268
2269 2269 mutex_enter(&door_knob);
2270 2270 /* is this door already on the unref list? */
2271 2271 if (d->door_flags & DOOR_UNREF_MULTI) {
2272 2272 door_node_t *dp;
2273 2273 for (dp = server->p_unref_list; dp != NULL;
2274 2274 dp = dp->door_ulist) {
2275 2275 if (d == dp) {
2276 2276 /* already there, don't need to add another */
2277 2277 mutex_exit(&door_knob);
2278 2278 VN_RELE(DTOV(d));
2279 2279 mutex_enter(&door_knob);
2280 2280 return;
2281 2281 }
2282 2282 }
2283 2283 }
2284 2284 ASSERT(d->door_ulist == NULL);
2285 2285 d->door_ulist = server->p_unref_list;
2286 2286 server->p_unref_list = d;
2287 2287 cv_broadcast(&server->p_unref_cv);
2288 2288 }
2289 2289
2290 2290 /*
2291 2291 * The callers buffer isn't big enough for all of the data/fd's. Allocate
2292 2292 * space in the callers address space for the results and copy the data
2293 2293 * there.
2294 2294 *
2295 2295 * For EOVERFLOW, we must clean up the server's door descriptors.
2296 2296 */
2297 2297 static int
2298 2298 door_overflow(
2299 2299 kthread_t *caller,
2300 2300 caddr_t data_ptr, /* data location */
2301 2301 size_t data_size, /* data size */
2302 2302 door_desc_t *desc_ptr, /* descriptor location */
2303 2303 uint_t desc_num) /* descriptor size */
2304 2304 {
2305 2305 proc_t *callerp = ttoproc(caller);
2306 2306 struct as *as = callerp->p_as;
2307 2307 door_client_t *ct = DOOR_CLIENT(caller->t_door);
2308 2308 caddr_t addr; /* Resulting address in target */
2309 2309 size_t rlen; /* Rounded len */
2310 2310 size_t len;
2311 2311 uint_t i;
2312 2312 size_t ds = desc_num * sizeof (door_desc_t);
2313 2313
2314 2314 ASSERT(MUTEX_NOT_HELD(&door_knob));
2315 2315 ASSERT(DOOR_T_HELD(ct) || ct->d_kernel);
2316 2316
2317 2317 /* Do initial overflow check */
2318 2318 if (!ufcanalloc(callerp, desc_num))
2319 2319 return (EMFILE);
2320 2320
2321 2321 /*
2322 2322 * Allocate space for this stuff in the callers address space
2323 2323 */
2324 2324 rlen = roundup(data_size + ds, PAGESIZE);
2325 2325 as_rangelock(as);
2326 2326 map_addr_proc(&addr, rlen, 0, 1, as->a_userlimit, ttoproc(caller), 0);
2327 2327 if (addr == NULL ||
2328 2328 as_map(as, addr, rlen, segvn_create, zfod_argsp) != 0) {
2329 2329 /* No virtual memory available, or anon mapping failed */
2330 2330 as_rangeunlock(as);
2331 2331 if (!ct->d_kernel && desc_num > 0) {
2332 2332 int error = door_release_fds(desc_ptr, desc_num);
2333 2333 if (error)
2334 2334 return (error);
2335 2335 }
2336 2336 return (EOVERFLOW);
2337 2337 }
2338 2338 as_rangeunlock(as);
2339 2339
2340 2340 if (ct->d_kernel)
2341 2341 goto out;
2342 2342
2343 2343 if (data_size != 0) {
2344 2344 caddr_t src = data_ptr;
2345 2345 caddr_t saddr = addr;
2346 2346
2347 2347 /* Copy any data */
2348 2348 len = data_size;
2349 2349 while (len != 0) {
2350 2350 int amount;
2351 2351 int error;
2352 2352
2353 2353 amount = len > PAGESIZE ? PAGESIZE : len;
2354 2354 if ((error = door_copy(as, src, saddr, amount)) != 0) {
2355 2355 (void) as_unmap(as, addr, rlen);
2356 2356 return (error);
2357 2357 }
2358 2358 saddr += amount;
2359 2359 src += amount;
2360 2360 len -= amount;
2361 2361 }
2362 2362 }
2363 2363 /* Copy any fd's */
2364 2364 if (desc_num != 0) {
2365 2365 door_desc_t *didpp, *start;
2366 2366 struct file **fpp;
2367 2367 int fpp_size;
2368 2368
2369 2369 start = didpp = kmem_alloc(ds, KM_SLEEP);
2370 2370 if (copyin_nowatch(desc_ptr, didpp, ds)) {
2371 2371 kmem_free(start, ds);
2372 2372 (void) as_unmap(as, addr, rlen);
2373 2373 return (EFAULT);
2374 2374 }
2375 2375
2376 2376 fpp_size = desc_num * sizeof (struct file *);
2377 2377 if (fpp_size > ct->d_fpp_size) {
2378 2378 /* make more space */
2379 2379 if (ct->d_fpp_size)
2380 2380 kmem_free(ct->d_fpp, ct->d_fpp_size);
2381 2381 ct->d_fpp_size = fpp_size;
2382 2382 ct->d_fpp = kmem_alloc(ct->d_fpp_size, KM_SLEEP);
2383 2383 }
2384 2384 fpp = ct->d_fpp;
2385 2385
2386 2386 for (i = 0; i < desc_num; i++) {
2387 2387 struct file *fp;
2388 2388 int fd = didpp->d_data.d_desc.d_descriptor;
2389 2389
2390 2390 if (!(didpp->d_attributes & DOOR_DESCRIPTOR) ||
2391 2391 (fp = getf(fd)) == NULL) {
2392 2392 /* close translated references */
2393 2393 door_fp_close(ct->d_fpp, fpp - ct->d_fpp);
2394 2394 /* close untranslated references */
2395 2395 door_fd_rele(didpp, desc_num - i, 0);
2396 2396 kmem_free(start, ds);
2397 2397 (void) as_unmap(as, addr, rlen);
2398 2398 return (EINVAL);
2399 2399 }
2400 2400 mutex_enter(&fp->f_tlock);
2401 2401 fp->f_count++;
2402 2402 mutex_exit(&fp->f_tlock);
2403 2403
2404 2404 *fpp = fp;
2405 2405 releasef(fd);
2406 2406
2407 2407 if (didpp->d_attributes & DOOR_RELEASE) {
2408 2408 /* release passed reference */
2409 2409 (void) closeandsetf(fd, NULL);
2410 2410 }
2411 2411
2412 2412 fpp++; didpp++;
2413 2413 }
2414 2414 kmem_free(start, ds);
2415 2415 }
2416 2416
2417 2417 out:
2418 2418 ct->d_overflow = 1;
2419 2419 ct->d_args.rbuf = addr;
2420 2420 ct->d_args.rsize = rlen;
2421 2421 return (0);
2422 2422 }
2423 2423
2424 2424 /*
2425 2425 * Transfer arguments from the client to the server.
2426 2426 */
2427 2427 static int
2428 2428 door_args(kthread_t *server, int is_private)
2429 2429 {
2430 2430 door_server_t *st = DOOR_SERVER(server->t_door);
2431 2431 door_client_t *ct = DOOR_CLIENT(curthread->t_door);
2432 2432 uint_t ndid;
2433 2433 size_t dsize;
2434 2434 int error;
2435 2435
2436 2436 ASSERT(DOOR_T_HELD(st));
2437 2437 ASSERT(MUTEX_NOT_HELD(&door_knob));
2438 2438
2439 2439 ndid = ct->d_args.desc_num;
2440 2440 if (ndid > door_max_desc)
2441 2441 return (E2BIG);
2442 2442
2443 2443 /*
2444 2444 * Get the stack layout, and fail now if it won't fit.
2445 2445 */
2446 2446 error = door_layout(server, ct->d_args.data_size, ndid, is_private);
2447 2447 if (error != 0)
2448 2448 return (error);
2449 2449
2450 2450 dsize = ndid * sizeof (door_desc_t);
2451 2451 if (ct->d_args.data_size != 0) {
2452 2452 if (ct->d_args.data_size <= door_max_arg) {
2453 2453 /*
2454 2454 * Use a 2 copy method for small amounts of data
2455 2455 *
2456 2456 * Allocate a little more than we need for the
2457 2457 * args, in the hope that the results will fit
2458 2458 * without having to reallocate a buffer
2459 2459 */
2460 2460 ASSERT(ct->d_buf == NULL);
2461 2461 ct->d_bufsize = roundup(ct->d_args.data_size,
2462 2462 DOOR_ROUND);
2463 2463 ct->d_buf = kmem_alloc(ct->d_bufsize, KM_SLEEP);
2464 2464 if (copyin_nowatch(ct->d_args.data_ptr,
2465 2465 ct->d_buf, ct->d_args.data_size) != 0) {
2466 2466 kmem_free(ct->d_buf, ct->d_bufsize);
2467 2467 ct->d_buf = NULL;
2468 2468 ct->d_bufsize = 0;
2469 2469 return (EFAULT);
2470 2470 }
2471 2471 } else {
2472 2472 struct as *as;
2473 2473 caddr_t src;
2474 2474 caddr_t dest;
2475 2475 size_t len = ct->d_args.data_size;
2476 2476 uintptr_t base;
2477 2477
2478 2478 /*
2479 2479 * Use a 1 copy method
2480 2480 */
2481 2481 as = ttoproc(server)->p_as;
2482 2482 src = ct->d_args.data_ptr;
2483 2483
2484 2484 dest = st->d_layout.dl_datap;
2485 2485 base = (uintptr_t)dest;
2486 2486
2487 2487 /*
2488 2488 * Copy data directly into server. We proceed
2489 2489 * downward from the top of the stack, to mimic
2490 2490 * normal stack usage. This allows the guard page
2491 2491 * to stop us before we corrupt anything.
2492 2492 */
2493 2493 while (len != 0) {
2494 2494 uintptr_t start;
2495 2495 uintptr_t end;
2496 2496 uintptr_t offset;
2497 2497 size_t amount;
2498 2498
2499 2499 /*
2500 2500 * Locate the next part to copy.
2501 2501 */
2502 2502 end = base + len;
2503 2503 start = P2ALIGN(end - 1, PAGESIZE);
2504 2504
2505 2505 /*
2506 2506 * if we are on the final (first) page, fix
2507 2507 * up the start position.
2508 2508 */
2509 2509 if (P2ALIGN(base, PAGESIZE) == start)
2510 2510 start = base;
2511 2511
2512 2512 offset = start - base; /* the copy offset */
2513 2513 amount = end - start; /* # bytes to copy */
2514 2514
2515 2515 ASSERT(amount > 0 && amount <= len &&
2516 2516 amount <= PAGESIZE);
2517 2517
2518 2518 error = door_copy(as, src + offset,
2519 2519 dest + offset, amount);
2520 2520 if (error != 0)
2521 2521 return (error);
2522 2522 len -= amount;
2523 2523 }
2524 2524 }
2525 2525 }
2526 2526 /*
2527 2527 * Copyin the door args and translate them into files
2528 2528 */
2529 2529 if (ndid != 0) {
2530 2530 door_desc_t *didpp;
2531 2531 door_desc_t *start;
2532 2532 struct file **fpp;
2533 2533
2534 2534 start = didpp = kmem_alloc(dsize, KM_SLEEP);
2535 2535
2536 2536 if (copyin_nowatch(ct->d_args.desc_ptr, didpp, dsize)) {
2537 2537 kmem_free(start, dsize);
2538 2538 return (EFAULT);
2539 2539 }
2540 2540 ct->d_fpp_size = ndid * sizeof (struct file *);
2541 2541 ct->d_fpp = kmem_alloc(ct->d_fpp_size, KM_SLEEP);
2542 2542 fpp = ct->d_fpp;
2543 2543 while (ndid--) {
2544 2544 struct file *fp;
2545 2545 int fd = didpp->d_data.d_desc.d_descriptor;
2546 2546
2547 2547 /* We only understand file descriptors as passed objs */
2548 2548 if (!(didpp->d_attributes & DOOR_DESCRIPTOR) ||
2549 2549 (fp = getf(fd)) == NULL) {
2550 2550 /* close translated references */
2551 2551 door_fp_close(ct->d_fpp, fpp - ct->d_fpp);
2552 2552 /* close untranslated references */
2553 2553 door_fd_rele(didpp, ndid + 1, 0);
2554 2554 kmem_free(start, dsize);
2555 2555 kmem_free(ct->d_fpp, ct->d_fpp_size);
2556 2556 ct->d_fpp = NULL;
2557 2557 ct->d_fpp_size = 0;
2558 2558 return (EINVAL);
2559 2559 }
2560 2560 /* Hold the fp */
2561 2561 mutex_enter(&fp->f_tlock);
2562 2562 fp->f_count++;
2563 2563 mutex_exit(&fp->f_tlock);
2564 2564
2565 2565 *fpp = fp;
2566 2566 releasef(fd);
2567 2567
2568 2568 if (didpp->d_attributes & DOOR_RELEASE) {
2569 2569 /* release passed reference */
2570 2570 (void) closeandsetf(fd, NULL);
2571 2571 }
2572 2572
2573 2573 fpp++; didpp++;
2574 2574 }
2575 2575 kmem_free(start, dsize);
2576 2576 }
2577 2577 return (0);
2578 2578 }
2579 2579
2580 2580 /*
2581 2581 * Transfer arguments from a user client to a kernel server. This copies in
2582 2582 * descriptors and translates them into door handles. It doesn't touch the
2583 2583 * other data, letting the kernel server deal with that (to avoid needing
2584 2584 * to copy the data twice).
2585 2585 */
2586 2586 static int
2587 2587 door_translate_in(void)
2588 2588 {
2589 2589 door_client_t *ct = DOOR_CLIENT(curthread->t_door);
2590 2590 uint_t ndid;
2591 2591
2592 2592 ASSERT(MUTEX_NOT_HELD(&door_knob));
2593 2593 ndid = ct->d_args.desc_num;
2594 2594 if (ndid > door_max_desc)
2595 2595 return (E2BIG);
2596 2596 /*
2597 2597 * Copyin the door args and translate them into door handles.
2598 2598 */
2599 2599 if (ndid != 0) {
2600 2600 door_desc_t *didpp;
2601 2601 door_desc_t *start;
2602 2602 size_t dsize = ndid * sizeof (door_desc_t);
2603 2603 struct file *fp;
2604 2604
2605 2605 start = didpp = kmem_alloc(dsize, KM_SLEEP);
2606 2606
2607 2607 if (copyin_nowatch(ct->d_args.desc_ptr, didpp, dsize)) {
2608 2608 kmem_free(start, dsize);
2609 2609 return (EFAULT);
2610 2610 }
2611 2611 while (ndid--) {
2612 2612 vnode_t *vp;
2613 2613 int fd = didpp->d_data.d_desc.d_descriptor;
2614 2614
2615 2615 /*
2616 2616 * We only understand file descriptors as passed objs
2617 2617 */
2618 2618 if ((didpp->d_attributes & DOOR_DESCRIPTOR) &&
2619 2619 (fp = getf(fd)) != NULL) {
2620 2620 didpp->d_data.d_handle = FTODH(fp);
2621 2621 /* Hold the door */
2622 2622 door_ki_hold(didpp->d_data.d_handle);
2623 2623
2624 2624 releasef(fd);
2625 2625
2626 2626 if (didpp->d_attributes & DOOR_RELEASE) {
2627 2627 /* release passed reference */
2628 2628 (void) closeandsetf(fd, NULL);
2629 2629 }
2630 2630
2631 2631 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
2632 2632 vp = fp->f_vnode;
2633 2633
2634 2634 /* Set attributes */
2635 2635 didpp->d_attributes = DOOR_HANDLE |
2636 2636 (VTOD(vp)->door_flags & DOOR_ATTR_MASK);
2637 2637 } else {
2638 2638 /* close translated references */
2639 2639 door_fd_close(start, didpp - start);
2640 2640 /* close untranslated references */
2641 2641 door_fd_rele(didpp, ndid + 1, 0);
2642 2642 kmem_free(start, dsize);
2643 2643 return (EINVAL);
2644 2644 }
2645 2645 didpp++;
2646 2646 }
2647 2647 ct->d_args.desc_ptr = start;
2648 2648 }
2649 2649 return (0);
2650 2650 }
2651 2651
2652 2652 /*
2653 2653 * Translate door arguments from kernel to user. This copies the passed
2654 2654 * door handles. It doesn't touch other data. It is used by door_upcall,
2655 2655 * and for data returned by a door_call to a kernel server.
2656 2656 */
2657 2657 static int
2658 2658 door_translate_out(void)
2659 2659 {
2660 2660 door_client_t *ct = DOOR_CLIENT(curthread->t_door);
2661 2661 uint_t ndid;
2662 2662
2663 2663 ASSERT(MUTEX_NOT_HELD(&door_knob));
2664 2664 ndid = ct->d_args.desc_num;
2665 2665 if (ndid > door_max_desc) {
2666 2666 door_fd_rele(ct->d_args.desc_ptr, ndid, 1);
2667 2667 return (E2BIG);
2668 2668 }
2669 2669 /*
2670 2670 * Translate the door args into files
2671 2671 */
2672 2672 if (ndid != 0) {
2673 2673 door_desc_t *didpp = ct->d_args.desc_ptr;
2674 2674 struct file **fpp;
2675 2675
2676 2676 ct->d_fpp_size = ndid * sizeof (struct file *);
2677 2677 fpp = ct->d_fpp = kmem_alloc(ct->d_fpp_size, KM_SLEEP);
2678 2678 while (ndid--) {
2679 2679 struct file *fp = NULL;
2680 2680 int fd = -1;
2681 2681
2682 2682 /*
2683 2683 * We understand file descriptors and door
2684 2684 * handles as passed objs.
2685 2685 */
2686 2686 if (didpp->d_attributes & DOOR_DESCRIPTOR) {
2687 2687 fd = didpp->d_data.d_desc.d_descriptor;
2688 2688 fp = getf(fd);
2689 2689 } else if (didpp->d_attributes & DOOR_HANDLE)
2690 2690 fp = DHTOF(didpp->d_data.d_handle);
2691 2691 if (fp != NULL) {
2692 2692 /* Hold the fp */
2693 2693 mutex_enter(&fp->f_tlock);
2694 2694 fp->f_count++;
2695 2695 mutex_exit(&fp->f_tlock);
2696 2696
2697 2697 *fpp = fp;
2698 2698 if (didpp->d_attributes & DOOR_DESCRIPTOR)
2699 2699 releasef(fd);
2700 2700 if (didpp->d_attributes & DOOR_RELEASE) {
2701 2701 /* release passed reference */
2702 2702 if (fd >= 0)
2703 2703 (void) closeandsetf(fd, NULL);
2704 2704 else
2705 2705 (void) closef(fp);
2706 2706 }
2707 2707 } else {
2708 2708 /* close translated references */
2709 2709 door_fp_close(ct->d_fpp, fpp - ct->d_fpp);
2710 2710 /* close untranslated references */
2711 2711 door_fd_rele(didpp, ndid + 1, 1);
2712 2712 kmem_free(ct->d_fpp, ct->d_fpp_size);
2713 2713 ct->d_fpp = NULL;
2714 2714 ct->d_fpp_size = 0;
2715 2715 return (EINVAL);
2716 2716 }
2717 2717 fpp++; didpp++;
2718 2718 }
2719 2719 }
2720 2720 return (0);
2721 2721 }
2722 2722
2723 2723 /*
2724 2724 * Move the results from the server to the client
2725 2725 */
2726 2726 static int
2727 2727 door_results(kthread_t *caller, caddr_t data_ptr, size_t data_size,
2728 2728 door_desc_t *desc_ptr, uint_t desc_num)
2729 2729 {
2730 2730 door_client_t *ct = DOOR_CLIENT(caller->t_door);
2731 2731 door_upcall_t *dup = ct->d_upcall;
2732 2732 size_t dsize;
2733 2733 size_t rlen;
2734 2734 size_t result_size;
2735 2735
2736 2736 ASSERT(DOOR_T_HELD(ct));
2737 2737 ASSERT(MUTEX_NOT_HELD(&door_knob));
2738 2738
2739 2739 if (ct->d_noresults)
2740 2740 return (E2BIG); /* No results expected */
2741 2741
2742 2742 if (desc_num > door_max_desc)
2743 2743 return (E2BIG); /* Too many descriptors */
2744 2744
2745 2745 dsize = desc_num * sizeof (door_desc_t);
2746 2746 /*
2747 2747 * Check if the results are bigger than the clients buffer
2748 2748 */
2749 2749 if (dsize)
2750 2750 rlen = roundup(data_size, sizeof (door_desc_t));
2751 2751 else
2752 2752 rlen = data_size;
2753 2753 if ((result_size = rlen + dsize) == 0)
2754 2754 return (0);
2755 2755
2756 2756 if (dup != NULL) {
2757 2757 if (desc_num > dup->du_max_descs)
2758 2758 return (EMFILE);
2759 2759
2760 2760 if (data_size > dup->du_max_data)
2761 2761 return (E2BIG);
2762 2762
2763 2763 /*
2764 2764 * Handle upcalls
2765 2765 */
2766 2766 if (ct->d_args.rbuf == NULL || ct->d_args.rsize < result_size) {
2767 2767 /*
2768 2768 * If there's no return buffer or the buffer is too
2769 2769 * small, allocate a new one. The old buffer (if it
2770 2770 * exists) will be freed by the upcall client.
2771 2771 */
2772 2772 if (result_size > door_max_upcall_reply)
2773 2773 return (E2BIG);
2774 2774 ct->d_args.rsize = result_size;
2775 2775 ct->d_args.rbuf = kmem_alloc(result_size, KM_SLEEP);
2776 2776 }
2777 2777 ct->d_args.data_ptr = ct->d_args.rbuf;
2778 2778 if (data_size != 0 &&
2779 2779 copyin_nowatch(data_ptr, ct->d_args.data_ptr,
2780 2780 data_size) != 0)
2781 2781 return (EFAULT);
2782 2782 } else if (result_size > ct->d_args.rsize) {
2783 2783 return (door_overflow(caller, data_ptr, data_size,
2784 2784 desc_ptr, desc_num));
2785 2785 } else if (data_size != 0) {
2786 2786 if (data_size <= door_max_arg) {
2787 2787 /*
2788 2788 * Use a 2 copy method for small amounts of data
2789 2789 */
2790 2790 if (ct->d_buf == NULL) {
2791 2791 ct->d_bufsize = data_size;
2792 2792 ct->d_buf = kmem_alloc(ct->d_bufsize, KM_SLEEP);
2793 2793 } else if (ct->d_bufsize < data_size) {
2794 2794 kmem_free(ct->d_buf, ct->d_bufsize);
2795 2795 ct->d_bufsize = data_size;
2796 2796 ct->d_buf = kmem_alloc(ct->d_bufsize, KM_SLEEP);
2797 2797 }
2798 2798 if (copyin_nowatch(data_ptr, ct->d_buf, data_size) != 0)
2799 2799 return (EFAULT);
2800 2800 } else {
2801 2801 struct as *as = ttoproc(caller)->p_as;
2802 2802 caddr_t dest = ct->d_args.rbuf;
2803 2803 caddr_t src = data_ptr;
2804 2804 size_t len = data_size;
2805 2805
2806 2806 /* Copy data directly into client */
2807 2807 while (len != 0) {
2808 2808 uint_t amount;
2809 2809 uint_t max;
2810 2810 uint_t off;
2811 2811 int error;
2812 2812
2813 2813 off = (uintptr_t)dest & PAGEOFFSET;
2814 2814 if (off)
2815 2815 max = PAGESIZE - off;
2816 2816 else
2817 2817 max = PAGESIZE;
2818 2818 amount = len > max ? max : len;
2819 2819 error = door_copy(as, src, dest, amount);
2820 2820 if (error != 0)
2821 2821 return (error);
2822 2822 dest += amount;
2823 2823 src += amount;
2824 2824 len -= amount;
2825 2825 }
2826 2826 }
2827 2827 }
2828 2828
2829 2829 /*
2830 2830 * Copyin the returned door ids and translate them into door_node_t
2831 2831 */
2832 2832 if (desc_num != 0) {
2833 2833 door_desc_t *start;
2834 2834 door_desc_t *didpp;
2835 2835 struct file **fpp;
2836 2836 size_t fpp_size;
2837 2837 uint_t i;
2838 2838
2839 2839 /* First, check if we would overflow client */
2840 2840 if (!ufcanalloc(ttoproc(caller), desc_num))
2841 2841 return (EMFILE);
2842 2842
2843 2843 start = didpp = kmem_alloc(dsize, KM_SLEEP);
2844 2844 if (copyin_nowatch(desc_ptr, didpp, dsize)) {
2845 2845 kmem_free(start, dsize);
2846 2846 return (EFAULT);
2847 2847 }
2848 2848 fpp_size = desc_num * sizeof (struct file *);
2849 2849 if (fpp_size > ct->d_fpp_size) {
2850 2850 /* make more space */
2851 2851 if (ct->d_fpp_size)
2852 2852 kmem_free(ct->d_fpp, ct->d_fpp_size);
2853 2853 ct->d_fpp_size = fpp_size;
2854 2854 ct->d_fpp = kmem_alloc(fpp_size, KM_SLEEP);
2855 2855 }
2856 2856 fpp = ct->d_fpp;
2857 2857
2858 2858 for (i = 0; i < desc_num; i++) {
2859 2859 struct file *fp;
2860 2860 int fd = didpp->d_data.d_desc.d_descriptor;
2861 2861
2862 2862 /* Only understand file descriptor results */
2863 2863 if (!(didpp->d_attributes & DOOR_DESCRIPTOR) ||
2864 2864 (fp = getf(fd)) == NULL) {
2865 2865 /* close translated references */
2866 2866 door_fp_close(ct->d_fpp, fpp - ct->d_fpp);
2867 2867 /* close untranslated references */
2868 2868 door_fd_rele(didpp, desc_num - i, 0);
2869 2869 kmem_free(start, dsize);
2870 2870 return (EINVAL);
2871 2871 }
2872 2872
2873 2873 mutex_enter(&fp->f_tlock);
2874 2874 fp->f_count++;
2875 2875 mutex_exit(&fp->f_tlock);
2876 2876
2877 2877 *fpp = fp;
2878 2878 releasef(fd);
2879 2879
2880 2880 if (didpp->d_attributes & DOOR_RELEASE) {
2881 2881 /* release passed reference */
2882 2882 (void) closeandsetf(fd, NULL);
2883 2883 }
2884 2884
2885 2885 fpp++; didpp++;
2886 2886 }
2887 2887 kmem_free(start, dsize);
2888 2888 }
2889 2889 return (0);
2890 2890 }
2891 2891
2892 2892 /*
2893 2893 * Close all the descriptors.
2894 2894 */
2895 2895 static void
2896 2896 door_fd_close(door_desc_t *d, uint_t n)
2897 2897 {
2898 2898 uint_t i;
2899 2899
2900 2900 ASSERT(MUTEX_NOT_HELD(&door_knob));
2901 2901 for (i = 0; i < n; i++) {
2902 2902 if (d->d_attributes & DOOR_DESCRIPTOR) {
2903 2903 (void) closeandsetf(
2904 2904 d->d_data.d_desc.d_descriptor, NULL);
2905 2905 } else if (d->d_attributes & DOOR_HANDLE) {
2906 2906 door_ki_rele(d->d_data.d_handle);
2907 2907 }
2908 2908 d++;
2909 2909 }
2910 2910 }
2911 2911
2912 2912 /*
2913 2913 * Close descriptors that have the DOOR_RELEASE attribute set.
2914 2914 */
2915 2915 void
2916 2916 door_fd_rele(door_desc_t *d, uint_t n, int from_kernel)
2917 2917 {
2918 2918 uint_t i;
2919 2919
2920 2920 ASSERT(MUTEX_NOT_HELD(&door_knob));
2921 2921 for (i = 0; i < n; i++) {
2922 2922 if (d->d_attributes & DOOR_RELEASE) {
2923 2923 if (d->d_attributes & DOOR_DESCRIPTOR) {
2924 2924 (void) closeandsetf(
2925 2925 d->d_data.d_desc.d_descriptor, NULL);
2926 2926 } else if (from_kernel &&
2927 2927 (d->d_attributes & DOOR_HANDLE)) {
2928 2928 door_ki_rele(d->d_data.d_handle);
2929 2929 }
2930 2930 }
2931 2931 d++;
2932 2932 }
2933 2933 }
2934 2934
2935 2935 /*
2936 2936 * Copy descriptors into the kernel so we can release any marked
2937 2937 * DOOR_RELEASE.
2938 2938 */
2939 2939 int
2940 2940 door_release_fds(door_desc_t *desc_ptr, uint_t ndesc)
2941 2941 {
2942 2942 size_t dsize;
2943 2943 door_desc_t *didpp;
2944 2944 uint_t desc_num;
2945 2945
2946 2946 ASSERT(MUTEX_NOT_HELD(&door_knob));
2947 2947 ASSERT(ndesc != 0);
2948 2948
2949 2949 desc_num = MIN(ndesc, door_max_desc);
2950 2950
2951 2951 dsize = desc_num * sizeof (door_desc_t);
2952 2952 didpp = kmem_alloc(dsize, KM_SLEEP);
2953 2953
2954 2954 while (ndesc > 0) {
2955 2955 uint_t count = MIN(ndesc, desc_num);
2956 2956
2957 2957 if (copyin_nowatch(desc_ptr, didpp,
2958 2958 count * sizeof (door_desc_t))) {
2959 2959 kmem_free(didpp, dsize);
2960 2960 return (EFAULT);
2961 2961 }
2962 2962 door_fd_rele(didpp, count, 0);
2963 2963
2964 2964 ndesc -= count;
2965 2965 desc_ptr += count;
2966 2966 }
2967 2967 kmem_free(didpp, dsize);
2968 2968 return (0);
2969 2969 }
2970 2970
2971 2971 /*
2972 2972 * Decrement ref count on all the files passed
2973 2973 */
2974 2974 static void
2975 2975 door_fp_close(struct file **fp, uint_t n)
2976 2976 {
2977 2977 uint_t i;
2978 2978
2979 2979 ASSERT(MUTEX_NOT_HELD(&door_knob));
2980 2980
2981 2981 for (i = 0; i < n; i++)
2982 2982 (void) closef(fp[i]);
2983 2983 }
2984 2984
2985 2985 /*
2986 2986 * Copy data from 'src' in current address space to 'dest' in 'as' for 'len'
2987 2987 * bytes.
2988 2988 *
2989 2989 * Performs this using 1 mapin and 1 copy operation.
2990 2990 *
2991 2991 * We really should do more than 1 page at a time to improve
2992 2992 * performance, but for now this is treated as an anomalous condition.
2993 2993 */
2994 2994 static int
2995 2995 door_copy(struct as *as, caddr_t src, caddr_t dest, uint_t len)
2996 2996 {
2997 2997 caddr_t kaddr;
2998 2998 caddr_t rdest;
2999 2999 uint_t off;
3000 3000 page_t **pplist;
3001 3001 page_t *pp = NULL;
3002 3002 int error = 0;
3003 3003
3004 3004 ASSERT(len <= PAGESIZE);
3005 3005 off = (uintptr_t)dest & PAGEOFFSET; /* offset within the page */
3006 3006 rdest = (caddr_t)((uintptr_t)dest &
3007 3007 (uintptr_t)PAGEMASK); /* Page boundary */
3008 3008 ASSERT(off + len <= PAGESIZE);
3009 3009
3010 3010 /*
3011 3011 * Lock down destination page.
3012 3012 */
3013 3013 if (as_pagelock(as, &pplist, rdest, PAGESIZE, S_WRITE))
3014 3014 return (E2BIG);
3015 3015 /*
3016 3016 * Check if we have a shadow page list from as_pagelock. If not,
3017 3017 * we took the slow path and have to find our page struct the hard
3018 3018 * way.
3019 3019 */
3020 3020 if (pplist == NULL) {
3021 3021 pfn_t pfnum;
3022 3022
3023 3023 /* MMU mapping is already locked down */
3024 3024 AS_LOCK_ENTER(as, RW_READER);
3025 3025 pfnum = hat_getpfnum(as->a_hat, rdest);
3026 3026 AS_LOCK_EXIT(as);
3027 3027
3028 3028 /*
3029 3029 * TODO: The pfn step should not be necessary - need
3030 3030 * a hat_getpp() function.
3031 3031 */
3032 3032 if (pf_is_memory(pfnum)) {
3033 3033 pp = page_numtopp_nolock(pfnum);
3034 3034 ASSERT(pp == NULL || PAGE_LOCKED(pp));
3035 3035 } else
3036 3036 pp = NULL;
3037 3037 if (pp == NULL) {
3038 3038 as_pageunlock(as, pplist, rdest, PAGESIZE, S_WRITE);
3039 3039 return (E2BIG);
3040 3040 }
3041 3041 } else {
3042 3042 pp = *pplist;
3043 3043 }
3044 3044 /*
3045 3045 * Map destination page into kernel address
3046 3046 */
3047 3047 if (kpm_enable)
3048 3048 kaddr = (caddr_t)hat_kpm_mapin(pp, (struct kpme *)NULL);
3049 3049 else
3050 3050 kaddr = (caddr_t)ppmapin(pp, PROT_READ | PROT_WRITE,
3051 3051 (caddr_t)-1);
3052 3052
3053 3053 /*
3054 3054 * Copy from src to dest
3055 3055 */
3056 3056 if (copyin_nowatch(src, kaddr + off, len) != 0)
3057 3057 error = EFAULT;
3058 3058 /*
3059 3059 * Unmap destination page from kernel
3060 3060 */
3061 3061 if (kpm_enable)
3062 3062 hat_kpm_mapout(pp, (struct kpme *)NULL, kaddr);
3063 3063 else
3064 3064 ppmapout(kaddr);
3065 3065 /*
3066 3066 * Unlock destination page
3067 3067 */
3068 3068 as_pageunlock(as, pplist, rdest, PAGESIZE, S_WRITE);
3069 3069 return (error);
3070 3070 }
3071 3071
3072 3072 /*
3073 3073 * General kernel upcall using doors
3074 3074 * Returns 0 on success, errno for failures.
3075 3075 * Caller must have a hold on the door based vnode, and on any
3076 3076 * references passed in desc_ptr. The references are released
3077 3077 * in the event of an error, and passed without duplication
3078 3078 * otherwise. Note that param->rbuf must be 64-bit aligned in
3079 3079 * a 64-bit kernel, since it may be used to store door descriptors
3080 3080 * if they are returned by the server. The caller is responsible
3081 3081 * for holding a reference to the cred passed in.
3082 3082 */
3083 3083 int
3084 3084 door_upcall(vnode_t *vp, door_arg_t *param, struct cred *cred,
3085 3085 size_t max_data, uint_t max_descs)
3086 3086 {
3087 3087 /* Locals */
3088 3088 door_upcall_t *dup;
3089 3089 door_node_t *dp;
3090 3090 kthread_t *server_thread;
3091 3091 int error = 0;
3092 3092 klwp_t *lwp;
3093 3093 door_client_t *ct; /* curthread door_data */
3094 3094 door_server_t *st; /* server thread door_data */
3095 3095 int gotresults = 0;
3096 3096 int cancel_pending;
3097 3097
3098 3098 if (vp->v_type != VDOOR) {
3099 3099 if (param->desc_num)
3100 3100 door_fd_rele(param->desc_ptr, param->desc_num, 1);
3101 3101 return (EINVAL);
3102 3102 }
3103 3103
3104 3104 lwp = ttolwp(curthread);
3105 3105 ct = door_my_client(1);
3106 3106 dp = VTOD(vp); /* Convert to a door_node_t */
3107 3107
3108 3108 dup = kmem_zalloc(sizeof (*dup), KM_SLEEP);
3109 3109 dup->du_cred = (cred != NULL) ? cred : curthread->t_cred;
3110 3110 dup->du_max_data = max_data;
3111 3111 dup->du_max_descs = max_descs;
3112 3112
3113 3113 /*
3114 3114 * This should be done in shuttle_resume(), just before going to
3115 3115 * sleep, but we want to avoid overhead while holding door_knob.
3116 3116 * prstop() is just a no-op if we don't really go to sleep.
3117 3117 * We test not-kernel-address-space for the sake of clustering code.
3118 3118 */
3119 3119 if (lwp && lwp->lwp_nostop == 0 && curproc->p_as != &kas)
3120 3120 prstop(PR_REQUESTED, 0);
3121 3121
3122 3122 mutex_enter(&door_knob);
3123 3123 if (DOOR_INVALID(dp)) {
3124 3124 mutex_exit(&door_knob);
3125 3125 if (param->desc_num)
3126 3126 door_fd_rele(param->desc_ptr, param->desc_num, 1);
3127 3127 error = EBADF;
3128 3128 goto out;
3129 3129 }
3130 3130
3131 3131 if (dp->door_target == &p0) {
3132 3132 /* Can't do an upcall to a kernel server */
3133 3133 mutex_exit(&door_knob);
3134 3134 if (param->desc_num)
3135 3135 door_fd_rele(param->desc_ptr, param->desc_num, 1);
3136 3136 error = EINVAL;
3137 3137 goto out;
3138 3138 }
3139 3139
3140 3140 error = door_check_limits(dp, param, 1);
3141 3141 if (error != 0) {
3142 3142 mutex_exit(&door_knob);
3143 3143 if (param->desc_num)
3144 3144 door_fd_rele(param->desc_ptr, param->desc_num, 1);
3145 3145 goto out;
3146 3146 }
3147 3147
3148 3148 /*
3149 3149 * Get a server thread from the target domain
3150 3150 */
3151 3151 if ((server_thread = door_get_server(dp)) == NULL) {
3152 3152 if (DOOR_INVALID(dp))
3153 3153 error = EBADF;
3154 3154 else
3155 3155 error = EAGAIN;
3156 3156 mutex_exit(&door_knob);
3157 3157 if (param->desc_num)
3158 3158 door_fd_rele(param->desc_ptr, param->desc_num, 1);
3159 3159 goto out;
3160 3160 }
3161 3161
3162 3162 st = DOOR_SERVER(server_thread->t_door);
3163 3163 ct->d_buf = param->data_ptr;
3164 3164 ct->d_bufsize = param->data_size;
3165 3165 ct->d_args = *param; /* structure assignment */
3166 3166
3167 3167 if (ct->d_args.desc_num) {
3168 3168 /*
3169 3169 * Move data from client to server
3170 3170 */
3171 3171 DOOR_T_HOLD(st);
3172 3172 mutex_exit(&door_knob);
3173 3173 error = door_translate_out();
3174 3174 mutex_enter(&door_knob);
3175 3175 DOOR_T_RELEASE(st);
3176 3176 if (error) {
3177 3177 /*
3178 3178 * We're not going to resume this thread after all
3179 3179 */
3180 3180 door_release_server(dp, server_thread);
3181 3181 shuttle_sleep(server_thread);
3182 3182 mutex_exit(&door_knob);
3183 3183 goto out;
3184 3184 }
3185 3185 }
3186 3186
3187 3187 ct->d_upcall = dup;
3188 3188 if (param->rsize == 0)
3189 3189 ct->d_noresults = 1;
3190 3190 else
3191 3191 ct->d_noresults = 0;
3192 3192
3193 3193 dp->door_active++;
3194 3194
3195 3195 ct->d_error = DOOR_WAIT;
3196 3196 st->d_caller = curthread;
3197 3197 st->d_active = dp;
3198 3198
3199 3199 shuttle_resume(server_thread, &door_knob);
3200 3200
3201 3201 mutex_enter(&door_knob);
3202 3202 shuttle_return:
3203 3203 if ((error = ct->d_error) < 0) { /* DOOR_WAIT or DOOR_EXIT */
3204 3204 /*
3205 3205 * Premature wakeup. Find out why (stop, forkall, sig, exit ...)
3206 3206 */
3207 3207 mutex_exit(&door_knob); /* May block in ISSIG */
3208 3208 cancel_pending = 0;
3209 3209 if (lwp && (ISSIG(curthread, FORREAL) || lwp->lwp_sysabort ||
3210 3210 MUSTRETURN(curproc, curthread) ||
3211 3211 (cancel_pending = schedctl_cancel_pending()) != 0)) {
3212 3212 /* Signal, forkall, ... */
3213 3213 if (cancel_pending)
3214 3214 schedctl_cancel_eintr();
3215 3215 lwp->lwp_sysabort = 0;
3216 3216 mutex_enter(&door_knob);
3217 3217 error = EINTR;
3218 3218 /*
3219 3219 * If the server has finished processing our call,
3220 3220 * or exited (calling door_slam()), then d_error
3221 3221 * will have changed. If the server hasn't finished
3222 3222 * yet, d_error will still be DOOR_WAIT, and we
3223 3223 * let it know we are not interested in any
3224 3224 * results by sending a SIGCANCEL, unless the door
3225 3225 * is marked with DOOR_NO_CANCEL.
3226 3226 */
3227 3227 if (ct->d_error == DOOR_WAIT &&
3228 3228 st->d_caller == curthread) {
3229 3229 proc_t *p = ttoproc(server_thread);
3230 3230
3231 3231 st->d_active = NULL;
3232 3232 st->d_caller = NULL;
3233 3233 if (!(dp->door_flags & DOOR_NO_CANCEL)) {
3234 3234 DOOR_T_HOLD(st);
3235 3235 mutex_exit(&door_knob);
3236 3236
3237 3237 mutex_enter(&p->p_lock);
3238 3238 sigtoproc(p, server_thread, SIGCANCEL);
3239 3239 mutex_exit(&p->p_lock);
3240 3240
3241 3241 mutex_enter(&door_knob);
3242 3242 DOOR_T_RELEASE(st);
3243 3243 }
3244 3244 }
3245 3245 } else {
3246 3246 /*
3247 3247 * Return from stop(), server exit...
3248 3248 *
3249 3249 * Note that the server could have done a
3250 3250 * door_return while the client was in stop state
3251 3251 * (ISSIG), in which case the error condition
3252 3252 * is updated by the server.
3253 3253 */
3254 3254 mutex_enter(&door_knob);
3255 3255 if (ct->d_error == DOOR_WAIT) {
3256 3256 /* Still waiting for a reply */
3257 3257 shuttle_swtch(&door_knob);
3258 3258 mutex_enter(&door_knob);
3259 3259 if (lwp)
3260 3260 lwp->lwp_asleep = 0;
3261 3261 goto shuttle_return;
3262 3262 } else if (ct->d_error == DOOR_EXIT) {
3263 3263 /* Server exit */
3264 3264 error = EINTR;
3265 3265 } else {
3266 3266 /* Server did a door_return during ISSIG */
3267 3267 error = ct->d_error;
3268 3268 }
3269 3269 }
3270 3270 /*
3271 3271 * Can't exit if the server is currently copying
3272 3272 * results for me
3273 3273 */
3274 3274 while (DOOR_T_HELD(ct))
3275 3275 cv_wait(&ct->d_cv, &door_knob);
3276 3276
3277 3277 /*
3278 3278 * Find out if results were successfully copied.
3279 3279 */
3280 3280 if (ct->d_error == 0)
3281 3281 gotresults = 1;
3282 3282 }
3283 3283 if (lwp) {
3284 3284 lwp->lwp_asleep = 0; /* /proc */
3285 3285 lwp->lwp_sysabort = 0; /* /proc */
3286 3286 }
3287 3287 if (--dp->door_active == 0 && (dp->door_flags & DOOR_DELAY))
3288 3288 door_deliver_unref(dp);
3289 3289 mutex_exit(&door_knob);
3290 3290
3291 3291 /*
3292 3292 * Translate returned doors (if any)
3293 3293 */
3294 3294
3295 3295 if (ct->d_noresults)
3296 3296 goto out;
3297 3297
3298 3298 if (error) {
3299 3299 /*
3300 3300 * If server returned results successfully, then we've
3301 3301 * been interrupted and may need to clean up.
3302 3302 */
3303 3303 if (gotresults) {
3304 3304 ASSERT(error == EINTR);
3305 3305 door_fp_close(ct->d_fpp, ct->d_args.desc_num);
3306 3306 }
3307 3307 goto out;
3308 3308 }
3309 3309
3310 3310 if (ct->d_args.desc_num) {
3311 3311 struct file **fpp;
3312 3312 door_desc_t *didpp;
3313 3313 vnode_t *vp;
3314 3314 uint_t n = ct->d_args.desc_num;
3315 3315
3316 3316 didpp = ct->d_args.desc_ptr = (door_desc_t *)(ct->d_args.rbuf +
3317 3317 roundup(ct->d_args.data_size, sizeof (door_desc_t)));
3318 3318 fpp = ct->d_fpp;
3319 3319
3320 3320 while (n--) {
3321 3321 struct file *fp;
3322 3322
3323 3323 fp = *fpp;
3324 3324 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
3325 3325 vp = fp->f_vnode;
3326 3326
3327 3327 didpp->d_attributes = DOOR_HANDLE |
3328 3328 (VTOD(vp)->door_flags & DOOR_ATTR_MASK);
3329 3329 didpp->d_data.d_handle = FTODH(fp);
3330 3330
3331 3331 fpp++; didpp++;
3332 3332 }
3333 3333 }
3334 3334
3335 3335 /* on return data is in rbuf */
3336 3336 *param = ct->d_args; /* structure assignment */
3337 3337
3338 3338 out:
3339 3339 kmem_free(dup, sizeof (*dup));
3340 3340
3341 3341 if (ct->d_fpp) {
3342 3342 kmem_free(ct->d_fpp, ct->d_fpp_size);
3343 3343 ct->d_fpp = NULL;
3344 3344 ct->d_fpp_size = 0;
3345 3345 }
3346 3346
3347 3347 ct->d_upcall = NULL;
3348 3348 ct->d_noresults = 0;
3349 3349 ct->d_buf = NULL;
3350 3350 ct->d_bufsize = 0;
3351 3351 return (error);
3352 3352 }
3353 3353
3354 3354 /*
3355 3355 * Add a door to the per-process list of active doors for which the
3356 3356 * process is a server.
3357 3357 */
3358 3358 static void
3359 3359 door_list_insert(door_node_t *dp)
3360 3360 {
3361 3361 proc_t *p = dp->door_target;
3362 3362
3363 3363 ASSERT(MUTEX_HELD(&door_knob));
3364 3364 dp->door_list = p->p_door_list;
3365 3365 p->p_door_list = dp;
3366 3366 }
3367 3367
3368 3368 /*
3369 3369 * Remove a door from the per-process list of active doors.
3370 3370 */
3371 3371 void
3372 3372 door_list_delete(door_node_t *dp)
3373 3373 {
3374 3374 door_node_t **pp;
3375 3375
3376 3376 ASSERT(MUTEX_HELD(&door_knob));
3377 3377 /*
3378 3378 * Find the door in the list. If the door belongs to another process,
3379 3379 * it's OK to use p_door_list since that process can't exit until all
3380 3380 * doors have been taken off the list (see door_exit).
3381 3381 */
3382 3382 pp = &(dp->door_target->p_door_list);
3383 3383 while (*pp != dp)
3384 3384 pp = &((*pp)->door_list);
3385 3385
3386 3386 /* found it, take it off the list */
3387 3387 *pp = dp->door_list;
3388 3388 }
3389 3389
3390 3390
3391 3391 /*
3392 3392 * External kernel interfaces for doors. These functions are available
3393 3393 * outside the doorfs module for use in creating and using doors from
3394 3394 * within the kernel.
3395 3395 */
3396 3396
3397 3397 /*
3398 3398 * door_ki_upcall invokes a user-level door server from the kernel, with
3399 3399 * the credentials associated with curthread.
3400 3400 */
3401 3401 int
3402 3402 door_ki_upcall(door_handle_t dh, door_arg_t *param)
3403 3403 {
3404 3404 return (door_ki_upcall_limited(dh, param, NULL, SIZE_MAX, UINT_MAX));
3405 3405 }
3406 3406
3407 3407 /*
3408 3408 * door_ki_upcall_limited invokes a user-level door server from the
3409 3409 * kernel with the given credentials and reply limits. If the "cred"
3410 3410 * argument is NULL, uses the credentials associated with current
3411 3411 * thread. max_data limits the maximum length of the returned data (the
3412 3412 * client will get E2BIG if they go over), and max_desc limits the
3413 3413 * number of returned descriptors (the client will get EMFILE if they
3414 3414 * go over).
3415 3415 */
3416 3416 int
3417 3417 door_ki_upcall_limited(door_handle_t dh, door_arg_t *param, struct cred *cred,
3418 3418 size_t max_data, uint_t max_desc)
3419 3419 {
3420 3420 file_t *fp = DHTOF(dh);
3421 3421 vnode_t *realvp;
3422 3422
3423 3423 if (VOP_REALVP(fp->f_vnode, &realvp, NULL))
3424 3424 realvp = fp->f_vnode;
3425 3425 return (door_upcall(realvp, param, cred, max_data, max_desc));
3426 3426 }
3427 3427
3428 3428 /*
3429 3429 * Function call to create a "kernel" door server. A kernel door
3430 3430 * server provides a way for a user-level process to invoke a function
3431 3431 * in the kernel through a door_call. From the caller's point of
3432 3432 * view, a kernel door server looks the same as a user-level one
3433 3433 * (except the server pid is 0). Unlike normal door calls, the
3434 3434 * kernel door function is invoked via a normal function call in the
3435 3435 * same thread and context as the caller.
3436 3436 */
3437 3437 int
3438 3438 door_ki_create(void (*pc_cookie)(), void *data_cookie, uint_t attributes,
3439 3439 door_handle_t *dhp)
3440 3440 {
3441 3441 int err;
3442 3442 file_t *fp;
3443 3443
3444 3444 /* no DOOR_PRIVATE */
3445 3445 if ((attributes & ~DOOR_KI_CREATE_MASK) ||
3446 3446 (attributes & (DOOR_UNREF | DOOR_UNREF_MULTI)) ==
3447 3447 (DOOR_UNREF | DOOR_UNREF_MULTI))
3448 3448 return (EINVAL);
3449 3449
3450 3450 err = door_create_common(pc_cookie, data_cookie, attributes,
3451 3451 1, NULL, &fp);
3452 3452 if (err == 0 && (attributes & (DOOR_UNREF | DOOR_UNREF_MULTI)) &&
3453 3453 p0.p_unref_thread == 0) {
3454 3454 /* need to create unref thread for process 0 */
3455 3455 (void) thread_create(NULL, 0, door_unref_kernel, NULL, 0, &p0,
3456 3456 TS_RUN, minclsyspri);
3457 3457 }
3458 3458 if (err == 0) {
3459 3459 *dhp = FTODH(fp);
3460 3460 }
3461 3461 return (err);
3462 3462 }
3463 3463
3464 3464 void
3465 3465 door_ki_hold(door_handle_t dh)
3466 3466 {
3467 3467 file_t *fp = DHTOF(dh);
3468 3468
3469 3469 mutex_enter(&fp->f_tlock);
3470 3470 fp->f_count++;
3471 3471 mutex_exit(&fp->f_tlock);
3472 3472 }
3473 3473
3474 3474 void
3475 3475 door_ki_rele(door_handle_t dh)
3476 3476 {
3477 3477 file_t *fp = DHTOF(dh);
3478 3478
3479 3479 (void) closef(fp);
3480 3480 }
3481 3481
3482 3482 int
3483 3483 door_ki_open(char *pathname, door_handle_t *dhp)
3484 3484 {
3485 3485 file_t *fp;
3486 3486 vnode_t *vp;
3487 3487 int err;
3488 3488
3489 3489 if ((err = lookupname(pathname, UIO_SYSSPACE, FOLLOW, NULL, &vp)) != 0)
3490 3490 return (err);
3491 3491 if (err = VOP_OPEN(&vp, FREAD, kcred, NULL)) {
3492 3492 VN_RELE(vp);
3493 3493 return (err);
3494 3494 }
3495 3495 if (vp->v_type != VDOOR) {
3496 3496 VN_RELE(vp);
3497 3497 return (EINVAL);
3498 3498 }
3499 3499 if ((err = falloc(vp, FREAD | FWRITE, &fp, NULL)) != 0) {
3500 3500 VN_RELE(vp);
3501 3501 return (err);
3502 3502 }
3503 3503 /* falloc returns with f_tlock held on success */
3504 3504 mutex_exit(&fp->f_tlock);
3505 3505 *dhp = FTODH(fp);
3506 3506 return (0);
3507 3507 }
3508 3508
3509 3509 int
3510 3510 door_ki_info(door_handle_t dh, struct door_info *dip)
3511 3511 {
3512 3512 file_t *fp = DHTOF(dh);
3513 3513 vnode_t *vp;
3514 3514
3515 3515 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
3516 3516 vp = fp->f_vnode;
3517 3517 if (vp->v_type != VDOOR)
3518 3518 return (EINVAL);
3519 3519 door_info_common(VTOD(vp), dip, fp);
3520 3520 return (0);
3521 3521 }
3522 3522
3523 3523 door_handle_t
3524 3524 door_ki_lookup(int did)
3525 3525 {
3526 3526 file_t *fp;
3527 3527 door_handle_t dh;
3528 3528
3529 3529 /* is the descriptor really a door? */
3530 3530 if (door_lookup(did, &fp) == NULL)
3531 3531 return (NULL);
3532 3532 /* got the door, put a hold on it and release the fd */
3533 3533 dh = FTODH(fp);
3534 3534 door_ki_hold(dh);
3535 3535 releasef(did);
3536 3536 return (dh);
3537 3537 }
3538 3538
3539 3539 int
3540 3540 door_ki_setparam(door_handle_t dh, int type, size_t val)
3541 3541 {
3542 3542 file_t *fp = DHTOF(dh);
3543 3543 vnode_t *vp;
3544 3544
3545 3545 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
3546 3546 vp = fp->f_vnode;
3547 3547 if (vp->v_type != VDOOR)
3548 3548 return (EINVAL);
3549 3549 return (door_setparam_common(VTOD(vp), 1, type, val));
3550 3550 }
3551 3551
3552 3552 int
3553 3553 door_ki_getparam(door_handle_t dh, int type, size_t *out)
3554 3554 {
3555 3555 file_t *fp = DHTOF(dh);
3556 3556 vnode_t *vp;
3557 3557
3558 3558 if (VOP_REALVP(fp->f_vnode, &vp, NULL))
3559 3559 vp = fp->f_vnode;
3560 3560 if (vp->v_type != VDOOR)
3561 3561 return (EINVAL);
3562 3562 return (door_getparam_common(VTOD(vp), type, out));
3563 3563 }
|
↓ open down ↓ |
3469 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX