Print this page
5253 kmem_alloc/kmem_zalloc won't fail with KM_SLEEP
5254 getrbuf won't fail with KM_SLEEP
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/avs/ns/sv/sv.c
+++ new/usr/src/uts/common/avs/ns/sv/sv.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 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 23 * Use is subject to license terms.
24 24 *
25 25 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
26 26 */
27 27
28 28 /*
29 29 * Storage Volume Character and Block Driver (SV)
30 30 *
31 31 * This driver implements a simplistic /dev/{r}dsk/ interface to a
32 32 * specified disk volume that is otherwise managed by the Prism
33 33 * software. The SV driver layers itself onto the underlying disk
34 34 * device driver by changing function pointers in the cb_ops
35 35 * structure.
36 36 *
37 37 * CONFIGURATION:
38 38 *
39 39 * 1. Configure the driver using the svadm utility.
40 40 * 2. Access the device as before through /dev/rdsk/c?t?d?s?
41 41 *
42 42 * LIMITATIONS:
43 43 *
44 44 * This driver should NOT be used to share a device between another
45 45 * DataServices user interface module (e.g., STE) and a user accessing
46 46 * the device through the block device in O_WRITE mode. This is because
47 47 * writes through the block device are asynchronous (due to the page
48 48 * cache) and so consistency between the block device user and the
49 49 * STE user cannot be guaranteed.
50 50 *
51 51 * Data is copied between system struct buf(9s) and nsc_vec_t. This is
52 52 * wasteful and slow.
53 53 */
54 54
55 55 #include <sys/debug.h>
56 56 #include <sys/types.h>
57 57
58 58 #include <sys/ksynch.h>
59 59 #include <sys/kmem.h>
60 60 #include <sys/errno.h>
61 61 #include <sys/varargs.h>
62 62 #include <sys/file.h>
63 63 #include <sys/open.h>
64 64 #include <sys/conf.h>
65 65 #include <sys/cred.h>
66 66 #include <sys/buf.h>
67 67 #include <sys/uio.h>
68 68 #ifndef DS_DDICT
69 69 #include <sys/pathname.h>
70 70 #endif
71 71 #include <sys/aio_req.h>
72 72 #include <sys/dkio.h>
73 73 #include <sys/vtoc.h>
74 74 #include <sys/cmn_err.h>
75 75 #include <sys/modctl.h>
76 76 #include <sys/ddi.h>
77 77 #include <sys/sunddi.h>
78 78 #include <sys/sunldi.h>
79 79 #include <sys/nsctl/nsvers.h>
80 80
81 81 #include <sys/nsc_thread.h>
82 82 #include <sys/unistat/spcs_s.h>
83 83 #include <sys/unistat/spcs_s_k.h>
84 84 #include <sys/unistat/spcs_errors.h>
85 85
86 86 #ifdef DS_DDICT
87 87 #include "../contract.h"
88 88 #endif
89 89
90 90 #include "../nsctl.h"
91 91
92 92
93 93 #include <sys/sdt.h> /* dtrace is S10 or later */
94 94
95 95 #include "sv.h"
96 96 #include "sv_impl.h"
97 97 #include "sv_efi.h"
98 98
99 99 #define MAX_EINTR_COUNT 1000
100 100
101 101 /*
102 102 * sv_mod_status
103 103 */
104 104 #define SV_PREVENT_UNLOAD 1
105 105 #define SV_ALLOW_UNLOAD 2
106 106
107 107 static const int sv_major_rev = ISS_VERSION_MAJ; /* Major number */
108 108 static const int sv_minor_rev = ISS_VERSION_MIN; /* Minor number */
109 109 static const int sv_micro_rev = ISS_VERSION_MIC; /* Micro number */
110 110 static const int sv_baseline_rev = ISS_VERSION_NUM; /* Baseline number */
111 111
112 112 #ifdef DKIOCPARTITION
113 113 /*
114 114 * CRC32 polynomial table needed for computing the checksums
115 115 * in an EFI vtoc.
116 116 */
117 117 static const uint32_t sv_crc32_table[256] = { CRC32_TABLE };
118 118 #endif
119 119
120 120 static clock_t sv_config_time; /* Time of successful {en,dis}able */
121 121 static int sv_debug; /* Set non-zero for debug to syslog */
122 122 static int sv_mod_status; /* Set to prevent modunload */
123 123
124 124 static dev_info_t *sv_dip; /* Single DIP for driver */
125 125 static kmutex_t sv_mutex; /* Protect global lists, etc. */
126 126
127 127 static nsc_mem_t *sv_mem; /* nsctl memory allocator token */
128 128
129 129
130 130 /*
131 131 * Per device and per major state.
132 132 */
133 133
134 134 #ifndef _SunOS_5_6
135 135 #define UNSAFE_ENTER()
136 136 #define UNSAFE_EXIT()
137 137 #else
138 138 #define UNSAFE_ENTER() mutex_enter(&unsafe_driver)
139 139 #define UNSAFE_EXIT() mutex_exit(&unsafe_driver)
140 140 #endif
141 141
142 142 /* hash table of major dev structures */
143 143 static sv_maj_t *sv_majors[SV_MAJOR_HASH_CNT] = {0};
144 144 static sv_dev_t *sv_devs; /* array of per device structures */
145 145 static int sv_max_devices; /* SV version of nsc_max_devices() */
146 146 static int sv_ndevices; /* number of SV enabled devices */
147 147
148 148 /*
149 149 * Threading.
150 150 */
151 151
152 152 int sv_threads_max = 1024; /* maximum # to dynamically alloc */
153 153 int sv_threads = 32; /* # to pre-allocate (see sv.conf) */
154 154 int sv_threads_extra = 0; /* addl # we would have alloc'ed */
155 155
156 156 static nstset_t *sv_tset; /* the threadset pointer */
157 157
158 158 static int sv_threads_hysteresis = 4; /* hysteresis for threadset resizing */
159 159 static int sv_threads_dev = 2; /* # of threads to alloc per device */
160 160 static int sv_threads_inc = 8; /* increment for changing the set */
161 161 static int sv_threads_needed; /* number of threads needed */
162 162 static int sv_no_threads; /* number of nsc_create errors */
163 163 static int sv_max_nlive; /* max number of threads running */
164 164
165 165
166 166
167 167 /*
168 168 * nsctl fd callbacks.
169 169 */
170 170
171 171 static int svattach_fd(blind_t);
172 172 static int svdetach_fd(blind_t);
173 173
174 174 static nsc_def_t sv_fd_def[] = {
175 175 { "Attach", (uintptr_t)svattach_fd, },
176 176 { "Detach", (uintptr_t)svdetach_fd, },
177 177 { 0, 0, }
178 178 };
179 179
180 180 /*
181 181 * cb_ops functions.
182 182 */
183 183
184 184 static int svopen(dev_t *, int, int, cred_t *);
185 185 static int svclose(dev_t, int, int, cred_t *);
186 186 static int svioctl(dev_t, int, intptr_t, int, cred_t *, int *);
187 187 static int svprint(dev_t, char *);
188 188
189 189 /*
190 190 * These next functions are layered into the underlying driver's devops.
191 191 */
192 192
193 193 static int sv_lyr_open(dev_t *, int, int, cred_t *);
194 194 static int sv_lyr_close(dev_t, int, int, cred_t *);
195 195 static int sv_lyr_strategy(struct buf *);
196 196 static int sv_lyr_read(dev_t, struct uio *, cred_t *);
197 197 static int sv_lyr_write(dev_t, struct uio *, cred_t *);
198 198 static int sv_lyr_aread(dev_t, struct aio_req *, cred_t *);
199 199 static int sv_lyr_awrite(dev_t, struct aio_req *, cred_t *);
200 200 static int sv_lyr_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
201 201
202 202 static struct cb_ops sv_cb_ops = {
203 203 svopen, /* open */
204 204 svclose, /* close */
205 205 nulldev, /* strategy */
206 206 svprint,
207 207 nodev, /* dump */
208 208 nodev, /* read */
209 209 nodev, /* write */
210 210 svioctl,
211 211 nodev, /* devmap */
212 212 nodev, /* mmap */
213 213 nodev, /* segmap */
214 214 nochpoll, /* poll */
215 215 ddi_prop_op,
216 216 NULL, /* NOT a stream */
217 217 D_NEW | D_MP | D_64BIT,
218 218 CB_REV,
219 219 nodev, /* aread */
220 220 nodev, /* awrite */
221 221 };
222 222
223 223
224 224 /*
225 225 * dev_ops functions.
226 226 */
227 227
228 228 static int sv_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
229 229 static int sv_attach(dev_info_t *, ddi_attach_cmd_t);
230 230 static int sv_detach(dev_info_t *, ddi_detach_cmd_t);
231 231
232 232 static struct dev_ops sv_ops = {
233 233 DEVO_REV,
234 234 0,
235 235 sv_getinfo,
236 236 nulldev, /* identify */
237 237 nulldev, /* probe */
238 238 sv_attach,
239 239 sv_detach,
240 240 nodev, /* reset */
241 241 &sv_cb_ops,
242 242 (struct bus_ops *)0
243 243 };
244 244
245 245 /*
246 246 * Module linkage.
247 247 */
248 248
249 249 extern struct mod_ops mod_driverops;
250 250
251 251 static struct modldrv modldrv = {
252 252 &mod_driverops,
253 253 "nws:Storage Volume:" ISS_VERSION_STR,
254 254 &sv_ops
255 255 };
256 256
257 257 static struct modlinkage modlinkage = {
258 258 MODREV_1,
259 259 &modldrv,
260 260 0
261 261 };
262 262
263 263
264 264 int
265 265 _init(void)
266 266 {
267 267 int error;
268 268
269 269 mutex_init(&sv_mutex, NULL, MUTEX_DRIVER, NULL);
270 270
271 271 if ((error = mod_install(&modlinkage)) != 0) {
272 272 mutex_destroy(&sv_mutex);
273 273 return (error);
274 274 }
275 275
276 276 #ifdef DEBUG
277 277 cmn_err(CE_CONT, "!sv (revision %d.%d.%d.%d, %s, %s)\n",
278 278 sv_major_rev, sv_minor_rev, sv_micro_rev, sv_baseline_rev,
279 279 ISS_VERSION_STR, BUILD_DATE_STR);
280 280 #else
281 281 if (sv_micro_rev) {
282 282 cmn_err(CE_CONT, "!sv (revision %d.%d.%d, %s, %s)\n",
283 283 sv_major_rev, sv_minor_rev, sv_micro_rev,
284 284 ISS_VERSION_STR, BUILD_DATE_STR);
285 285 } else {
286 286 cmn_err(CE_CONT, "!sv (revision %d.%d, %s, %s)\n",
287 287 sv_major_rev, sv_minor_rev,
288 288 ISS_VERSION_STR, BUILD_DATE_STR);
289 289 }
290 290 #endif
291 291
292 292 return (error);
293 293 }
294 294
295 295
296 296 int
297 297 _fini(void)
298 298 {
299 299 int error;
300 300
301 301 if ((error = mod_remove(&modlinkage)) != 0)
302 302 return (error);
303 303
304 304 mutex_destroy(&sv_mutex);
305 305
306 306 return (error);
307 307 }
308 308
309 309
310 310 int
311 311 _info(struct modinfo *modinfop)
312 312 {
313 313 return (mod_info(&modlinkage, modinfop));
314 314 }
315 315
316 316
317 317 /*
318 318 * Locking & State.
319 319 *
320 320 * sv_mutex protects config information - sv_maj_t and sv_dev_t lists;
321 321 * threadset creation and sizing; sv_ndevices.
322 322 *
323 323 * If we need to hold both sv_mutex and sv_lock, then the sv_mutex
324 324 * must be acquired first.
325 325 *
326 326 * sv_lock protects the sv_dev_t structure for an individual device.
327 327 *
328 328 * sv_olock protects the otyp/open members of the sv_dev_t. If we need
329 329 * to hold both sv_lock and sv_olock, then the sv_lock must be acquired
330 330 * first.
331 331 *
332 332 * nsc_reserve/nsc_release are used in NSC_MULTI mode to allow multiple
333 333 * I/O operations to a device simultaneously, as above.
334 334 *
335 335 * All nsc_open/nsc_close/nsc_reserve/nsc_release operations that occur
336 336 * with sv_lock write-locked must be done with (sv_state == SV_PENDING)
337 337 * and (sv_pending == curthread) so that any recursion through
338 338 * sv_lyr_open/sv_lyr_close can be detected.
339 339 */
340 340
341 341
342 342 static int
343 343 sv_init_devs(void)
344 344 {
345 345 int i;
346 346
347 347 ASSERT(MUTEX_HELD(&sv_mutex));
348 348
349 349 if (sv_max_devices > 0)
350 350 return (0);
351 351
352 352 sv_max_devices = nsc_max_devices();
353 353
354 354 if (sv_max_devices <= 0) {
355 355 /* nsctl is not attached (nskernd not running) */
356 356 if (sv_debug > 0)
357 357 cmn_err(CE_CONT, "!sv: nsc_max_devices = 0\n");
358 358 return (EAGAIN);
359 359 }
360 360
361 361 sv_devs = nsc_kmem_zalloc((sv_max_devices * sizeof (*sv_devs)),
362 362 KM_NOSLEEP, sv_mem);
363 363
364 364 if (sv_devs == NULL) {
365 365 cmn_err(CE_WARN, "!sv: could not allocate sv_devs array");
366 366 return (ENOMEM);
367 367 }
368 368
369 369 for (i = 0; i < sv_max_devices; i++) {
370 370 mutex_init(&sv_devs[i].sv_olock, NULL, MUTEX_DRIVER, NULL);
371 371 rw_init(&sv_devs[i].sv_lock, NULL, RW_DRIVER, NULL);
372 372 }
373 373
374 374 if (sv_debug > 0)
375 375 cmn_err(CE_CONT, "!sv: sv_init_devs successful\n");
376 376
377 377 return (0);
378 378 }
379 379
380 380
381 381 static int
382 382 sv_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
383 383 {
384 384 int rc;
385 385
386 386 switch (cmd) {
387 387
388 388 case DDI_ATTACH:
389 389 sv_dip = dip;
390 390
391 391 if (ddi_create_minor_node(dip, "sv", S_IFCHR,
392 392 0, DDI_PSEUDO, 0) != DDI_SUCCESS)
393 393 goto failed;
394 394
395 395 mutex_enter(&sv_mutex);
396 396
397 397 sv_mem = nsc_register_mem("SV", NSC_MEM_LOCAL, 0);
398 398 if (sv_mem == NULL) {
399 399 mutex_exit(&sv_mutex);
400 400 goto failed;
401 401 }
402 402
403 403 rc = sv_init_devs();
404 404 if (rc != 0 && rc != EAGAIN) {
405 405 mutex_exit(&sv_mutex);
406 406 goto failed;
407 407 }
408 408
409 409 mutex_exit(&sv_mutex);
410 410
411 411
412 412 ddi_report_dev(dip);
413 413
414 414 sv_threads = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
415 415 DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
416 416 "sv_threads", sv_threads);
417 417
418 418 if (sv_debug > 0)
419 419 cmn_err(CE_CONT, "!sv: sv_threads=%d\n", sv_threads);
420 420
421 421 if (sv_threads > sv_threads_max)
422 422 sv_threads_max = sv_threads;
423 423
424 424 return (DDI_SUCCESS);
425 425
426 426 default:
427 427 return (DDI_FAILURE);
428 428 }
429 429
430 430 failed:
431 431 DTRACE_PROBE(sv_attach_failed);
432 432 (void) sv_detach(dip, DDI_DETACH);
433 433 return (DDI_FAILURE);
434 434 }
435 435
436 436
437 437 static int
438 438 sv_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
439 439 {
440 440 sv_dev_t *svp;
441 441 int i;
442 442
443 443 switch (cmd) {
444 444
445 445 case DDI_DETACH:
446 446
447 447 /*
448 448 * Check that everything is disabled.
449 449 */
450 450
451 451 mutex_enter(&sv_mutex);
452 452
453 453 if (sv_mod_status == SV_PREVENT_UNLOAD) {
454 454 mutex_exit(&sv_mutex);
455 455 DTRACE_PROBE(sv_detach_err_prevent);
456 456 return (DDI_FAILURE);
457 457 }
458 458
459 459 for (i = 0; sv_devs && i < sv_max_devices; i++) {
460 460 svp = &sv_devs[i];
461 461
462 462 if (svp->sv_state != SV_DISABLE) {
463 463 mutex_exit(&sv_mutex);
464 464 DTRACE_PROBE(sv_detach_err_busy);
465 465 return (DDI_FAILURE);
466 466 }
467 467 }
468 468
469 469
470 470 for (i = 0; sv_devs && i < sv_max_devices; i++) {
471 471 mutex_destroy(&sv_devs[i].sv_olock);
472 472 rw_destroy(&sv_devs[i].sv_lock);
473 473 }
474 474
475 475 if (sv_devs) {
476 476 nsc_kmem_free(sv_devs,
477 477 (sv_max_devices * sizeof (*sv_devs)));
478 478 sv_devs = NULL;
479 479 }
480 480 sv_max_devices = 0;
481 481
482 482 if (sv_mem) {
483 483 nsc_unregister_mem(sv_mem);
484 484 sv_mem = NULL;
485 485 }
486 486
487 487 mutex_exit(&sv_mutex);
488 488
489 489 /*
490 490 * Remove all minor nodes.
491 491 */
492 492
493 493 ddi_remove_minor_node(dip, NULL);
494 494 sv_dip = NULL;
495 495
496 496 return (DDI_SUCCESS);
497 497
498 498 default:
499 499 return (DDI_FAILURE);
500 500 }
501 501 }
502 502
503 503 static sv_maj_t *
504 504 sv_getmajor(const dev_t dev)
505 505 {
506 506 sv_maj_t **insert, *maj;
507 507 major_t umaj = getmajor(dev);
508 508
509 509 /*
510 510 * See if the hash table entry, or one of the hash chains
511 511 * is already allocated for this major number
512 512 */
513 513 if ((maj = sv_majors[SV_MAJOR_HASH(umaj)]) != 0) {
514 514 do {
515 515 if (maj->sm_major == umaj)
516 516 return (maj);
517 517 } while ((maj = maj->sm_next) != 0);
518 518 }
519 519
520 520 /*
521 521 * If the sv_mutex is held, there is design flaw, as the only non-mutex
522 522 * held callers can be sv_enable() or sv_dev_to_sv()
523 523 * Return an error, instead of panicing the system
524 524 */
525 525 if (MUTEX_HELD(&sv_mutex)) {
526 526 cmn_err(CE_WARN, "!sv: could not allocate sv_maj_t");
527 527 return (NULL);
528 528 }
529 529
530 530 /*
531 531 * Determine where to allocate a new element in the hash table
532 532 */
533 533 mutex_enter(&sv_mutex);
534 534 insert = &(sv_majors[SV_MAJOR_HASH(umaj)]);
535 535 for (maj = *insert; maj; maj = maj->sm_next) {
536 536
537 537 /* Did another thread beat us to it? */
538 538 if (maj->sm_major == umaj)
539 539 return (maj);
540 540
541 541 /* Find a NULL insert point? */
542 542 if (maj->sm_next == NULL)
543 543 insert = &maj->sm_next;
544 544 }
545 545
546 546 /*
547 547 * Located the new insert point
548 548 */
549 549 *insert = nsc_kmem_zalloc(sizeof (*maj), KM_NOSLEEP, sv_mem);
550 550 if ((maj = *insert) != 0)
551 551 maj->sm_major = umaj;
552 552 else
553 553 cmn_err(CE_WARN, "!sv: could not allocate sv_maj_t");
554 554
555 555 mutex_exit(&sv_mutex);
556 556
557 557 return (maj);
558 558 }
559 559
560 560 /* ARGSUSED */
561 561
562 562 static int
563 563 sv_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
564 564 {
565 565 int rc = DDI_FAILURE;
566 566
567 567 switch (infocmd) {
568 568
569 569 case DDI_INFO_DEVT2DEVINFO:
570 570 *result = sv_dip;
571 571 rc = DDI_SUCCESS;
572 572 break;
573 573
574 574 case DDI_INFO_DEVT2INSTANCE:
575 575 /*
576 576 * We only have a single instance.
577 577 */
578 578 *result = 0;
579 579 rc = DDI_SUCCESS;
580 580 break;
581 581
582 582 default:
583 583 break;
584 584 }
585 585
586 586 return (rc);
587 587 }
588 588
589 589
590 590 /*
591 591 * Hashing of devices onto major device structures.
592 592 *
593 593 * Individual device structures are hashed onto one of the sm_hash[]
594 594 * buckets in the relevant major device structure.
595 595 *
596 596 * Hash insertion and deletion -must- be done with sv_mutex held. Hash
597 597 * searching does not require the mutex because of the sm_seq member.
598 598 * sm_seq is incremented on each insertion (-after- hash chain pointer
599 599 * manipulation) and each deletion (-before- hash chain pointer
600 600 * manipulation). When searching the hash chain, the seq number is
601 601 * checked before accessing each device structure, if the seq number has
602 602 * changed, then we restart the search from the top of the hash chain.
603 603 * If we restart more than SV_HASH_RETRY times, we take sv_mutex and search
604 604 * the hash chain (we are guaranteed that this search cannot be
605 605 * interrupted).
606 606 */
607 607
608 608 #define SV_HASH_RETRY 16
609 609
610 610 static sv_dev_t *
611 611 sv_dev_to_sv(const dev_t dev, sv_maj_t **majpp)
612 612 {
613 613 minor_t umin = getminor(dev);
614 614 sv_dev_t **hb, *next, *svp;
615 615 sv_maj_t *maj;
616 616 int seq;
617 617 int try;
618 618
619 619 /* Get major hash table */
620 620 maj = sv_getmajor(dev);
621 621 if (majpp)
622 622 *majpp = maj;
623 623 if (maj == NULL)
624 624 return (NULL);
625 625
626 626 if (maj->sm_inuse == 0) {
627 627 DTRACE_PROBE1(
628 628 sv_dev_to_sv_end,
629 629 dev_t, dev);
630 630 return (NULL);
631 631 }
632 632
633 633 hb = &(maj->sm_hash[SV_MINOR_HASH(umin)]);
634 634 try = 0;
635 635
636 636 retry:
637 637 if (try > SV_HASH_RETRY)
638 638 mutex_enter(&sv_mutex);
639 639
640 640 seq = maj->sm_seq;
641 641 for (svp = *hb; svp; svp = next) {
642 642 next = svp->sv_hash;
643 643
644 644 nsc_membar_stld(); /* preserve register load order */
645 645
646 646 if (maj->sm_seq != seq) {
647 647 DTRACE_PROBE1(sv_dev_to_sv_retry, dev_t, dev);
648 648 try++;
649 649 goto retry;
650 650 }
651 651
652 652 if (svp->sv_dev == dev)
653 653 break;
654 654 }
655 655
656 656 if (try > SV_HASH_RETRY)
657 657 mutex_exit(&sv_mutex);
658 658
659 659 return (svp);
660 660 }
661 661
662 662
663 663 /*
664 664 * Must be called with sv_mutex held.
665 665 */
666 666
667 667 static int
668 668 sv_get_state(const dev_t udev, sv_dev_t **svpp)
669 669 {
670 670 sv_dev_t **hb, **insert, *svp;
671 671 sv_maj_t *maj;
672 672 minor_t umin;
673 673 int i;
674 674
675 675 /* Get major hash table */
676 676 if ((maj = sv_getmajor(udev)) == NULL)
677 677 return (NULL);
678 678
679 679 /* Determine which minor hash table */
680 680 umin = getminor(udev);
681 681 hb = &(maj->sm_hash[SV_MINOR_HASH(umin)]);
682 682
683 683 /* look for clash */
684 684
685 685 insert = hb;
686 686
687 687 for (svp = *hb; svp; svp = svp->sv_hash) {
688 688 if (svp->sv_dev == udev)
689 689 break;
690 690
691 691 if (svp->sv_hash == NULL)
692 692 insert = &svp->sv_hash;
693 693 }
694 694
695 695 if (svp) {
696 696 DTRACE_PROBE1(
697 697 sv_get_state_enabled,
698 698 dev_t, udev);
699 699 return (SV_EENABLED);
700 700 }
701 701
702 702 /* look for spare sv_devs slot */
703 703
704 704 for (i = 0; i < sv_max_devices; i++) {
705 705 svp = &sv_devs[i];
706 706
707 707 if (svp->sv_state == SV_DISABLE)
708 708 break;
709 709 }
710 710
711 711 if (i >= sv_max_devices) {
712 712 DTRACE_PROBE1(
713 713 sv_get_state_noslots,
714 714 dev_t, udev);
715 715 return (SV_ENOSLOTS);
716 716 }
717 717
718 718 svp->sv_state = SV_PENDING;
719 719 svp->sv_pending = curthread;
720 720
721 721 *insert = svp;
722 722 svp->sv_hash = NULL;
723 723 maj->sm_seq++; /* must be after the store to the hash chain */
724 724
725 725 *svpp = svp;
726 726
727 727 /*
728 728 * We do not know the size of the underlying device at
729 729 * this stage, so initialise "nblocks" property to
730 730 * zero, and update it whenever we succeed in
731 731 * nsc_reserve'ing the underlying nsc_fd_t.
732 732 */
733 733
734 734 svp->sv_nblocks = 0;
735 735
736 736 return (0);
737 737 }
738 738
739 739
740 740 /*
741 741 * Remove a device structure from it's hash chain.
742 742 * Must be called with sv_mutex held.
743 743 */
744 744
745 745 static void
746 746 sv_rm_hash(sv_dev_t *svp)
747 747 {
748 748 sv_dev_t **svpp;
749 749 sv_maj_t *maj;
750 750
751 751 /* Get major hash table */
752 752 if ((maj = sv_getmajor(svp->sv_dev)) == NULL)
753 753 return;
754 754
755 755 /* remove svp from hash chain */
756 756
757 757 svpp = &(maj->sm_hash[SV_MINOR_HASH(getminor(svp->sv_dev))]);
758 758 while (*svpp) {
759 759 if (*svpp == svp) {
760 760 /*
761 761 * increment of sm_seq must be before the
762 762 * removal from the hash chain
763 763 */
764 764 maj->sm_seq++;
765 765 *svpp = svp->sv_hash;
766 766 break;
767 767 }
768 768
769 769 svpp = &(*svpp)->sv_hash;
770 770 }
771 771
772 772 svp->sv_hash = NULL;
773 773 }
774 774
775 775 /*
776 776 * Free (disable) a device structure.
777 777 * Must be called with sv_lock(RW_WRITER) and sv_mutex held, and will
778 778 * perform the exits during its processing.
779 779 */
780 780
781 781 static int
782 782 sv_free(sv_dev_t *svp, const int error)
783 783 {
784 784 struct cb_ops *cb_ops;
785 785 sv_maj_t *maj;
786 786
787 787 /* Get major hash table */
788 788 if ((maj = sv_getmajor(svp->sv_dev)) == NULL)
789 789 return (NULL);
790 790
791 791 svp->sv_state = SV_PENDING;
792 792 svp->sv_pending = curthread;
793 793
794 794 /*
795 795 * Close the fd's before removing from the hash or swapping
796 796 * back the cb_ops pointers so that the cache flushes before new
797 797 * io can come in.
798 798 */
799 799
800 800 if (svp->sv_fd) {
801 801 (void) nsc_close(svp->sv_fd);
802 802 svp->sv_fd = 0;
803 803 }
804 804
805 805 sv_rm_hash(svp);
806 806
807 807 if (error != SV_ESDOPEN &&
808 808 error != SV_ELYROPEN && --maj->sm_inuse == 0) {
809 809
810 810 if (maj->sm_dev_ops)
811 811 cb_ops = maj->sm_dev_ops->devo_cb_ops;
812 812 else
813 813 cb_ops = NULL;
814 814
815 815 if (cb_ops && maj->sm_strategy != NULL) {
816 816 cb_ops->cb_strategy = maj->sm_strategy;
817 817 cb_ops->cb_close = maj->sm_close;
818 818 cb_ops->cb_ioctl = maj->sm_ioctl;
819 819 cb_ops->cb_write = maj->sm_write;
820 820 cb_ops->cb_open = maj->sm_open;
821 821 cb_ops->cb_read = maj->sm_read;
822 822 cb_ops->cb_flag = maj->sm_flag;
823 823
824 824 if (maj->sm_awrite)
825 825 cb_ops->cb_awrite = maj->sm_awrite;
826 826
827 827 if (maj->sm_aread)
828 828 cb_ops->cb_aread = maj->sm_aread;
829 829
830 830 /*
831 831 * corbin XXX
832 832 * Leave backing device ops in maj->sm_*
833 833 * to handle any requests that might come
834 834 * in during the disable. This could be
835 835 * a problem however if the backing device
836 836 * driver is changed while we process these
837 837 * requests.
838 838 *
839 839 * maj->sm_strategy = 0;
840 840 * maj->sm_awrite = 0;
841 841 * maj->sm_write = 0;
842 842 * maj->sm_ioctl = 0;
843 843 * maj->sm_close = 0;
844 844 * maj->sm_aread = 0;
845 845 * maj->sm_read = 0;
846 846 * maj->sm_open = 0;
847 847 * maj->sm_flag = 0;
848 848 *
849 849 */
850 850 }
851 851
852 852 if (maj->sm_dev_ops) {
853 853 maj->sm_dev_ops = 0;
854 854 }
855 855 }
856 856
857 857 if (svp->sv_lh) {
858 858 cred_t *crp = ddi_get_cred();
859 859
860 860 /*
861 861 * Close the protective layered driver open using the
862 862 * Sun Private layered driver i/f.
863 863 */
864 864
865 865 (void) ldi_close(svp->sv_lh, FREAD|FWRITE, crp);
866 866 svp->sv_lh = NULL;
867 867 }
868 868
869 869 svp->sv_timestamp = nsc_lbolt();
870 870 svp->sv_state = SV_DISABLE;
871 871 svp->sv_pending = NULL;
872 872 rw_exit(&svp->sv_lock);
873 873 mutex_exit(&sv_mutex);
874 874
875 875 return (error);
876 876 }
877 877
878 878 /*
879 879 * Reserve the device, taking into account the possibility that
880 880 * the reserve might have to be retried.
881 881 */
882 882 static int
883 883 sv_reserve(nsc_fd_t *fd, int flags)
884 884 {
885 885 int eintr_count;
886 886 int rc;
887 887
888 888 eintr_count = 0;
889 889 do {
890 890 rc = nsc_reserve(fd, flags);
891 891 if (rc == EINTR) {
892 892 ++eintr_count;
893 893 delay(2);
894 894 }
895 895 } while ((rc == EINTR) && (eintr_count < MAX_EINTR_COUNT));
896 896
897 897 return (rc);
898 898 }
899 899
900 900 static int
901 901 sv_enable(const caddr_t path, const int flag,
902 902 const dev_t udev, spcs_s_info_t kstatus)
903 903 {
904 904 struct dev_ops *dev_ops;
905 905 struct cb_ops *cb_ops;
906 906 sv_dev_t *svp;
907 907 sv_maj_t *maj;
908 908 nsc_size_t nblocks;
909 909 int rc;
910 910 cred_t *crp;
911 911 ldi_ident_t li;
912 912
913 913 if (udev == (dev_t)-1 || udev == 0) {
914 914 DTRACE_PROBE1(
915 915 sv_enable_err_baddev,
916 916 dev_t, udev);
917 917 return (SV_EBADDEV);
918 918 }
919 919
920 920 if ((flag & ~(NSC_CACHE|NSC_DEVICE)) != 0) {
921 921 DTRACE_PROBE1(sv_enable_err_amode, dev_t, udev);
922 922 return (SV_EAMODE);
923 923 }
924 924
925 925 /* Get major hash table */
926 926 if ((maj = sv_getmajor(udev)) == NULL)
927 927 return (SV_EBADDEV);
928 928
929 929 mutex_enter(&sv_mutex);
930 930
931 931 rc = sv_get_state(udev, &svp);
932 932 if (rc) {
933 933 mutex_exit(&sv_mutex);
934 934 DTRACE_PROBE1(sv_enable_err_state, dev_t, udev);
935 935 return (rc);
936 936 }
937 937
938 938 rw_enter(&svp->sv_lock, RW_WRITER);
939 939
940 940 /*
941 941 * Get real fd used for io
942 942 */
943 943
944 944 svp->sv_dev = udev;
945 945 svp->sv_flag = flag;
946 946
947 947 /*
948 948 * OR in NSC_DEVICE to ensure that nskern grabs the real strategy
949 949 * function pointer before sv swaps them out.
950 950 */
951 951
952 952 svp->sv_fd = nsc_open(path, (svp->sv_flag | NSC_DEVICE),
953 953 sv_fd_def, (blind_t)udev, &rc);
954 954
955 955 if (svp->sv_fd == NULL) {
956 956 if (kstatus)
957 957 spcs_s_add(kstatus, rc);
958 958 DTRACE_PROBE1(sv_enable_err_fd, dev_t, udev);
959 959 return (sv_free(svp, SV_ESDOPEN));
960 960 }
961 961
962 962 /*
963 963 * Perform a layered driver open using the Sun Private layered
964 964 * driver i/f to ensure that the cb_ops structure for the driver
965 965 * is not detached out from under us whilst sv is enabled.
966 966 *
967 967 */
968 968
969 969 crp = ddi_get_cred();
970 970 svp->sv_lh = NULL;
971 971
972 972 if ((rc = ldi_ident_from_dev(svp->sv_dev, &li)) == 0) {
973 973 rc = ldi_open_by_dev(&svp->sv_dev,
974 974 OTYP_BLK, FREAD|FWRITE, crp, &svp->sv_lh, li);
975 975 }
976 976
977 977 if (rc != 0) {
978 978 if (kstatus)
979 979 spcs_s_add(kstatus, rc);
980 980 DTRACE_PROBE1(sv_enable_err_lyr_open, dev_t, udev);
981 981 return (sv_free(svp, SV_ELYROPEN));
982 982 }
983 983
984 984 /*
985 985 * Do layering if required - must happen after nsc_open().
986 986 */
987 987
988 988 if (maj->sm_inuse++ == 0) {
989 989 maj->sm_dev_ops = nsc_get_devops(getmajor(udev));
990 990
991 991 if (maj->sm_dev_ops == NULL ||
992 992 maj->sm_dev_ops->devo_cb_ops == NULL) {
993 993 DTRACE_PROBE1(sv_enable_err_load, dev_t, udev);
994 994 return (sv_free(svp, SV_ELOAD));
995 995 }
996 996
997 997 dev_ops = maj->sm_dev_ops;
998 998 cb_ops = dev_ops->devo_cb_ops;
999 999
1000 1000 if (cb_ops->cb_strategy == NULL ||
1001 1001 cb_ops->cb_strategy == nodev ||
1002 1002 cb_ops->cb_strategy == nulldev) {
1003 1003 DTRACE_PROBE1(sv_enable_err_nostrategy, dev_t, udev);
1004 1004 return (sv_free(svp, SV_ELOAD));
1005 1005 }
1006 1006
1007 1007 if (cb_ops->cb_strategy == sv_lyr_strategy) {
1008 1008 DTRACE_PROBE1(sv_enable_err_svstrategy, dev_t, udev);
1009 1009 return (sv_free(svp, SV_ESTRATEGY));
1010 1010 }
1011 1011
1012 1012 maj->sm_strategy = cb_ops->cb_strategy;
1013 1013 maj->sm_close = cb_ops->cb_close;
1014 1014 maj->sm_ioctl = cb_ops->cb_ioctl;
1015 1015 maj->sm_write = cb_ops->cb_write;
1016 1016 maj->sm_open = cb_ops->cb_open;
1017 1017 maj->sm_read = cb_ops->cb_read;
1018 1018 maj->sm_flag = cb_ops->cb_flag;
1019 1019
1020 1020 cb_ops->cb_flag = cb_ops->cb_flag | D_MP;
1021 1021 cb_ops->cb_strategy = sv_lyr_strategy;
1022 1022 cb_ops->cb_close = sv_lyr_close;
1023 1023 cb_ops->cb_ioctl = sv_lyr_ioctl;
1024 1024 cb_ops->cb_write = sv_lyr_write;
1025 1025 cb_ops->cb_open = sv_lyr_open;
1026 1026 cb_ops->cb_read = sv_lyr_read;
1027 1027
1028 1028 /*
1029 1029 * Check that the driver has async I/O entry points
1030 1030 * before changing them.
1031 1031 */
1032 1032
1033 1033 if (dev_ops->devo_rev < 3 || cb_ops->cb_rev < 1) {
1034 1034 maj->sm_awrite = 0;
1035 1035 maj->sm_aread = 0;
1036 1036 } else {
1037 1037 maj->sm_awrite = cb_ops->cb_awrite;
1038 1038 maj->sm_aread = cb_ops->cb_aread;
1039 1039
1040 1040 cb_ops->cb_awrite = sv_lyr_awrite;
1041 1041 cb_ops->cb_aread = sv_lyr_aread;
1042 1042 }
1043 1043
1044 1044 /*
1045 1045 * Bug 4645743
1046 1046 *
1047 1047 * Prevent sv from ever unloading after it has interposed
1048 1048 * on a major device because there is a race between
1049 1049 * sv removing its layered entry points from the target
1050 1050 * dev_ops, a client coming in and accessing the driver,
1051 1051 * and the kernel modunloading the sv text.
1052 1052 *
1053 1053 * To allow unload, do svboot -u, which only happens in
1054 1054 * pkgrm time.
1055 1055 */
1056 1056 ASSERT(MUTEX_HELD(&sv_mutex));
1057 1057 sv_mod_status = SV_PREVENT_UNLOAD;
1058 1058 }
1059 1059
1060 1060
1061 1061 svp->sv_timestamp = nsc_lbolt();
1062 1062 svp->sv_state = SV_ENABLE;
1063 1063 svp->sv_pending = NULL;
1064 1064 rw_exit(&svp->sv_lock);
1065 1065
1066 1066 sv_ndevices++;
1067 1067 mutex_exit(&sv_mutex);
1068 1068
1069 1069 nblocks = 0;
1070 1070 if (sv_reserve(svp->sv_fd, NSC_READ|NSC_MULTI|NSC_PCATCH) == 0) {
1071 1071 nblocks = svp->sv_nblocks;
1072 1072 nsc_release(svp->sv_fd);
1073 1073 }
1074 1074
1075 1075 cmn_err(CE_CONT, "!sv: rdev 0x%lx, nblocks %" NSC_SZFMT "\n",
1076 1076 svp->sv_dev, nblocks);
1077 1077
1078 1078 return (0);
1079 1079 }
1080 1080
1081 1081
1082 1082 static int
1083 1083 sv_prepare_unload()
1084 1084 {
1085 1085 int rc = 0;
1086 1086
1087 1087 mutex_enter(&sv_mutex);
1088 1088
1089 1089 if (sv_mod_status == SV_PREVENT_UNLOAD) {
1090 1090 if ((sv_ndevices != 0) || (sv_tset != NULL)) {
1091 1091 rc = EBUSY;
1092 1092 } else {
1093 1093 sv_mod_status = SV_ALLOW_UNLOAD;
1094 1094 delay(SV_WAIT_UNLOAD * drv_usectohz(1000000));
1095 1095 }
1096 1096 }
1097 1097
1098 1098 mutex_exit(&sv_mutex);
1099 1099 return (rc);
1100 1100 }
1101 1101
1102 1102 static int
1103 1103 svattach_fd(blind_t arg)
1104 1104 {
1105 1105 dev_t dev = (dev_t)arg;
1106 1106 sv_dev_t *svp = sv_dev_to_sv(dev, NULL);
1107 1107 int rc;
1108 1108
1109 1109 if (sv_debug > 0)
1110 1110 cmn_err(CE_CONT, "!svattach_fd(%p, %p)\n", arg, (void *)svp);
1111 1111
1112 1112 if (svp == NULL) {
1113 1113 cmn_err(CE_WARN, "!svattach_fd: no state (arg %p)", arg);
1114 1114 return (0);
1115 1115 }
1116 1116
1117 1117 if ((rc = nsc_partsize(svp->sv_fd, &svp->sv_nblocks)) != 0) {
1118 1118 cmn_err(CE_WARN,
1119 1119 "!svattach_fd: nsc_partsize() failed, rc %d", rc);
1120 1120 svp->sv_nblocks = 0;
1121 1121 }
1122 1122
1123 1123 if ((rc = nsc_maxfbas(svp->sv_fd, 0, &svp->sv_maxfbas)) != 0) {
1124 1124 cmn_err(CE_WARN,
1125 1125 "!svattach_fd: nsc_maxfbas() failed, rc %d", rc);
1126 1126 svp->sv_maxfbas = 0;
1127 1127 }
1128 1128
1129 1129 if (sv_debug > 0) {
1130 1130 cmn_err(CE_CONT,
1131 1131 "!svattach_fd(%p): size %" NSC_SZFMT ", "
1132 1132 "maxfbas %" NSC_SZFMT "\n",
1133 1133 arg, svp->sv_nblocks, svp->sv_maxfbas);
1134 1134 }
1135 1135
1136 1136 return (0);
1137 1137 }
1138 1138
1139 1139
1140 1140 static int
1141 1141 svdetach_fd(blind_t arg)
1142 1142 {
1143 1143 dev_t dev = (dev_t)arg;
1144 1144 sv_dev_t *svp = sv_dev_to_sv(dev, NULL);
1145 1145
1146 1146 if (sv_debug > 0)
1147 1147 cmn_err(CE_CONT, "!svdetach_fd(%p, %p)\n", arg, (void *)svp);
1148 1148
1149 1149 /* svp can be NULL during disable of an sv */
1150 1150 if (svp == NULL)
1151 1151 return (0);
1152 1152
1153 1153 svp->sv_maxfbas = 0;
1154 1154 svp->sv_nblocks = 0;
1155 1155 return (0);
1156 1156 }
1157 1157
1158 1158
1159 1159 /*
1160 1160 * Side effect: if called with (guard != 0), then expects both sv_mutex
1161 1161 * and sv_lock(RW_WRITER) to be held, and will release them before returning.
1162 1162 */
1163 1163
1164 1164 /* ARGSUSED */
1165 1165 static int
1166 1166 sv_disable(dev_t dev, spcs_s_info_t kstatus)
1167 1167 {
1168 1168 sv_dev_t *svp = sv_dev_to_sv(dev, NULL);
1169 1169
1170 1170 if (svp == NULL) {
1171 1171
1172 1172 DTRACE_PROBE1(sv_disable_err_nodev, sv_dev_t *, svp);
1173 1173 return (SV_ENODEV);
1174 1174 }
1175 1175
1176 1176 mutex_enter(&sv_mutex);
1177 1177 rw_enter(&svp->sv_lock, RW_WRITER);
1178 1178
1179 1179 if (svp->sv_fd == NULL || svp->sv_state != SV_ENABLE) {
1180 1180 rw_exit(&svp->sv_lock);
1181 1181 mutex_exit(&sv_mutex);
1182 1182
1183 1183 DTRACE_PROBE1(sv_disable_err_disabled, sv_dev_t *, svp);
1184 1184 return (SV_EDISABLED);
1185 1185 }
1186 1186
1187 1187
1188 1188 sv_ndevices--;
1189 1189 return (sv_free(svp, 0));
1190 1190 }
1191 1191
1192 1192
1193 1193
1194 1194 static int
1195 1195 sv_lyr_open(dev_t *devp, int flag, int otyp, cred_t *crp)
1196 1196 {
1197 1197 nsc_buf_t *tmph;
1198 1198 sv_dev_t *svp;
1199 1199 sv_maj_t *maj;
1200 1200 int (*fn)();
1201 1201 dev_t odev;
1202 1202 int ret;
1203 1203 int rc;
1204 1204
1205 1205 svp = sv_dev_to_sv(*devp, &maj);
1206 1206
1207 1207 if (svp) {
1208 1208 if (svp->sv_state == SV_PENDING &&
1209 1209 svp->sv_pending == curthread) {
1210 1210 /*
1211 1211 * This is a recursive open from a call to
1212 1212 * ddi_lyr_open_by_devt and so we just want
1213 1213 * to pass it straight through to the
1214 1214 * underlying driver.
1215 1215 */
1216 1216 DTRACE_PROBE2(sv_lyr_open_recursive,
1217 1217 sv_dev_t *, svp,
1218 1218 dev_t, *devp);
1219 1219 svp = NULL;
1220 1220 } else
1221 1221 rw_enter(&svp->sv_lock, RW_READER);
1222 1222 }
1223 1223
1224 1224 odev = *devp;
1225 1225
1226 1226 if (maj && (fn = maj->sm_open) != 0) {
1227 1227 if (!(maj->sm_flag & D_MP)) {
1228 1228 UNSAFE_ENTER();
1229 1229 ret = (*fn)(devp, flag, otyp, crp);
1230 1230 UNSAFE_EXIT();
1231 1231 } else {
1232 1232 ret = (*fn)(devp, flag, otyp, crp);
1233 1233 }
1234 1234
1235 1235 if (ret == 0) {
1236 1236 /*
1237 1237 * Re-acquire svp if the driver changed *devp.
1238 1238 */
1239 1239
1240 1240 if (*devp != odev) {
1241 1241 if (svp != NULL)
1242 1242 rw_exit(&svp->sv_lock);
1243 1243
1244 1244 svp = sv_dev_to_sv(*devp, NULL);
1245 1245
1246 1246 if (svp) {
1247 1247 rw_enter(&svp->sv_lock, RW_READER);
1248 1248 }
1249 1249 }
1250 1250 }
1251 1251 } else {
1252 1252 ret = ENODEV;
1253 1253 }
1254 1254
1255 1255 if (svp && ret != 0 && svp->sv_state == SV_ENABLE) {
1256 1256 /*
1257 1257 * Underlying DDI open failed, but we have this
1258 1258 * device SV enabled. If we can read some data
1259 1259 * from the device, fake a successful open (this
1260 1260 * probably means that this device is RDC'd and we
1261 1261 * are getting the data from the secondary node).
1262 1262 *
1263 1263 * The reserve must be done with NSC_TRY|NSC_NOWAIT to
1264 1264 * ensure that it does not deadlock if this open is
1265 1265 * coming from nskernd:get_bsize().
1266 1266 */
1267 1267 rc = sv_reserve(svp->sv_fd,
1268 1268 NSC_TRY | NSC_NOWAIT | NSC_MULTI | NSC_PCATCH);
1269 1269 if (rc == 0) {
1270 1270 tmph = NULL;
1271 1271
1272 1272 rc = nsc_alloc_buf(svp->sv_fd, 0, 1, NSC_READ, &tmph);
1273 1273 if (rc <= 0) {
1274 1274 /* success */
1275 1275 ret = 0;
1276 1276 }
1277 1277
1278 1278 if (tmph) {
1279 1279 (void) nsc_free_buf(tmph);
1280 1280 tmph = NULL;
1281 1281 }
1282 1282
1283 1283 nsc_release(svp->sv_fd);
1284 1284
1285 1285 /*
1286 1286 * Count the number of layered opens that we
1287 1287 * fake since we have to fake a matching number
1288 1288 * of closes (OTYP_LYR open/close calls must be
1289 1289 * paired).
1290 1290 */
1291 1291
1292 1292 if (ret == 0 && otyp == OTYP_LYR) {
1293 1293 mutex_enter(&svp->sv_olock);
1294 1294 svp->sv_openlcnt++;
1295 1295 mutex_exit(&svp->sv_olock);
1296 1296 }
1297 1297 }
1298 1298 }
1299 1299
1300 1300 if (svp) {
1301 1301 rw_exit(&svp->sv_lock);
1302 1302 }
1303 1303
1304 1304 return (ret);
1305 1305 }
1306 1306
1307 1307
1308 1308 static int
1309 1309 sv_lyr_close(dev_t dev, int flag, int otyp, cred_t *crp)
1310 1310 {
1311 1311 sv_dev_t *svp;
1312 1312 sv_maj_t *maj;
1313 1313 int (*fn)();
1314 1314 int ret;
1315 1315
1316 1316 svp = sv_dev_to_sv(dev, &maj);
1317 1317
1318 1318 if (svp &&
1319 1319 svp->sv_state == SV_PENDING &&
1320 1320 svp->sv_pending == curthread) {
1321 1321 /*
1322 1322 * This is a recursive open from a call to
1323 1323 * ddi_lyr_close and so we just want
1324 1324 * to pass it straight through to the
1325 1325 * underlying driver.
1326 1326 */
1327 1327 DTRACE_PROBE2(sv_lyr_close_recursive, sv_dev_t *, svp,
1328 1328 dev_t, dev);
1329 1329 svp = NULL;
1330 1330 }
1331 1331
1332 1332 if (svp) {
1333 1333 rw_enter(&svp->sv_lock, RW_READER);
1334 1334
1335 1335 if (otyp == OTYP_LYR) {
1336 1336 mutex_enter(&svp->sv_olock);
1337 1337
1338 1338 if (svp->sv_openlcnt) {
1339 1339 /*
1340 1340 * Consume sufficient layered closes to
1341 1341 * account for the opens that we faked
1342 1342 * whilst the device was failed.
1343 1343 */
1344 1344 svp->sv_openlcnt--;
1345 1345 mutex_exit(&svp->sv_olock);
1346 1346 rw_exit(&svp->sv_lock);
1347 1347
1348 1348 DTRACE_PROBE1(sv_lyr_close_end, dev_t, dev);
1349 1349
1350 1350 return (0);
1351 1351 }
1352 1352
1353 1353 mutex_exit(&svp->sv_olock);
1354 1354 }
1355 1355 }
1356 1356
1357 1357 if (maj && (fn = maj->sm_close) != 0) {
1358 1358 if (!(maj->sm_flag & D_MP)) {
1359 1359 UNSAFE_ENTER();
1360 1360 ret = (*fn)(dev, flag, otyp, crp);
1361 1361 UNSAFE_EXIT();
1362 1362 } else {
1363 1363 ret = (*fn)(dev, flag, otyp, crp);
1364 1364 }
1365 1365 } else {
1366 1366 ret = ENODEV;
1367 1367 }
1368 1368
1369 1369 if (svp) {
1370 1370 rw_exit(&svp->sv_lock);
1371 1371 }
1372 1372
1373 1373 return (ret);
1374 1374 }
1375 1375
1376 1376
1377 1377 /*
1378 1378 * Convert the specified dev_t into a locked and enabled sv_dev_t, or
1379 1379 * return NULL.
1380 1380 */
1381 1381 static sv_dev_t *
1382 1382 sv_find_enabled(const dev_t dev, sv_maj_t **majpp)
1383 1383 {
1384 1384 sv_dev_t *svp;
1385 1385
1386 1386 while ((svp = sv_dev_to_sv(dev, majpp)) != NULL) {
1387 1387 rw_enter(&svp->sv_lock, RW_READER);
1388 1388
1389 1389 if (svp->sv_state == SV_ENABLE) {
1390 1390 /* locked and enabled */
1391 1391 break;
1392 1392 }
1393 1393
1394 1394 /*
1395 1395 * State was changed while waiting on the lock.
1396 1396 * Wait for a stable state.
1397 1397 */
1398 1398 rw_exit(&svp->sv_lock);
1399 1399
1400 1400 DTRACE_PROBE1(sv_find_enabled_retry, dev_t, dev);
1401 1401
1402 1402 delay(2);
1403 1403 }
1404 1404
1405 1405 return (svp);
1406 1406 }
1407 1407
1408 1408
1409 1409 static int
1410 1410 sv_lyr_uio(dev_t dev, uio_t *uiop, cred_t *crp, int rw)
1411 1411 {
1412 1412 sv_dev_t *svp;
1413 1413 sv_maj_t *maj;
1414 1414 int (*fn)();
1415 1415 int rc;
1416 1416
1417 1417 svp = sv_find_enabled(dev, &maj);
1418 1418 if (svp == NULL) {
1419 1419 if (maj) {
1420 1420 if (rw == NSC_READ)
1421 1421 fn = maj->sm_read;
1422 1422 else
1423 1423 fn = maj->sm_write;
1424 1424
1425 1425 if (fn != 0) {
1426 1426 if (!(maj->sm_flag & D_MP)) {
1427 1427 UNSAFE_ENTER();
1428 1428 rc = (*fn)(dev, uiop, crp);
1429 1429 UNSAFE_EXIT();
1430 1430 } else {
1431 1431 rc = (*fn)(dev, uiop, crp);
1432 1432 }
1433 1433 }
1434 1434
1435 1435 return (rc);
1436 1436 } else {
1437 1437 return (ENODEV);
1438 1438 }
1439 1439 }
1440 1440
1441 1441 ASSERT(RW_READ_HELD(&svp->sv_lock));
1442 1442
1443 1443 if (svp->sv_flag == 0) {
1444 1444 /*
1445 1445 * guard access mode
1446 1446 * - prevent user level access to the device
1447 1447 */
1448 1448 DTRACE_PROBE1(sv_lyr_uio_err_guard, uio_t *, uiop);
1449 1449 rc = EPERM;
1450 1450 goto out;
1451 1451 }
1452 1452
1453 1453 if ((rc = sv_reserve(svp->sv_fd, NSC_MULTI|NSC_PCATCH)) != 0) {
1454 1454 DTRACE_PROBE1(sv_lyr_uio_err_rsrv, uio_t *, uiop);
1455 1455 goto out;
1456 1456 }
1457 1457
1458 1458 if (rw == NSC_READ)
1459 1459 rc = nsc_uread(svp->sv_fd, uiop, crp);
1460 1460 else
1461 1461 rc = nsc_uwrite(svp->sv_fd, uiop, crp);
1462 1462
1463 1463 nsc_release(svp->sv_fd);
1464 1464
1465 1465 out:
1466 1466 rw_exit(&svp->sv_lock);
1467 1467
1468 1468 return (rc);
1469 1469 }
1470 1470
1471 1471
1472 1472 static int
1473 1473 sv_lyr_read(dev_t dev, uio_t *uiop, cred_t *crp)
1474 1474 {
1475 1475 return (sv_lyr_uio(dev, uiop, crp, NSC_READ));
1476 1476 }
1477 1477
1478 1478
1479 1479 static int
1480 1480 sv_lyr_write(dev_t dev, uio_t *uiop, cred_t *crp)
1481 1481 {
1482 1482 return (sv_lyr_uio(dev, uiop, crp, NSC_WRITE));
1483 1483 }
1484 1484
1485 1485
1486 1486 /* ARGSUSED */
1487 1487
1488 1488 static int
1489 1489 sv_lyr_aread(dev_t dev, struct aio_req *aio, cred_t *crp)
1490 1490 {
1491 1491 return (aphysio(sv_lyr_strategy,
1492 1492 anocancel, dev, B_READ, minphys, aio));
1493 1493 }
1494 1494
1495 1495
1496 1496 /* ARGSUSED */
1497 1497
1498 1498 static int
1499 1499 sv_lyr_awrite(dev_t dev, struct aio_req *aio, cred_t *crp)
1500 1500 {
1501 1501 return (aphysio(sv_lyr_strategy,
1502 1502 anocancel, dev, B_WRITE, minphys, aio));
1503 1503 }
1504 1504
1505 1505
1506 1506 /*
1507 1507 * Set up an array containing the list of raw path names
1508 1508 * The array for the paths is svl and the size of the array is
1509 1509 * in size.
1510 1510 *
1511 1511 * If there are more layered devices than will fit in the array,
1512 1512 * the number of extra layered devices is returned. Otherwise
1513 1513 * zero is return.
1514 1514 *
1515 1515 * Input:
1516 1516 * svn : array for paths
1517 1517 * size : size of the array
1518 1518 *
1519 1519 * Output (extra):
1520 1520 * zero : All paths fit in array
1521 1521 * >0 : Number of defined layered devices don't fit in array
1522 1522 */
1523 1523
1524 1524 static int
1525 1525 sv_list(void *ptr, const int size, int *extra, const int ilp32)
1526 1526 {
1527 1527 sv_name32_t *svn32;
1528 1528 sv_name_t *svn;
1529 1529 sv_dev_t *svp;
1530 1530 int *mode, *nblocks;
1531 1531 int i, index;
1532 1532 char *path;
1533 1533
1534 1534 *extra = 0;
1535 1535 index = 0;
1536 1536
1537 1537 if (ilp32)
1538 1538 svn32 = ptr;
1539 1539 else
1540 1540 svn = ptr;
1541 1541
1542 1542 mutex_enter(&sv_mutex);
1543 1543 for (i = 0; i < sv_max_devices; i++) {
1544 1544 svp = &sv_devs[i];
1545 1545
1546 1546 rw_enter(&svp->sv_lock, RW_READER);
1547 1547
1548 1548 if (svp->sv_state != SV_ENABLE) {
1549 1549 rw_exit(&svp->sv_lock);
1550 1550 continue;
1551 1551 }
1552 1552
1553 1553 if ((*extra) != 0 || ptr == NULL) {
1554 1554 /* Another overflow entry */
1555 1555 rw_exit(&svp->sv_lock);
1556 1556 (*extra)++;
1557 1557 continue;
1558 1558 }
1559 1559
1560 1560 if (ilp32) {
1561 1561 nblocks = &svn32->svn_nblocks;
1562 1562 mode = &svn32->svn_mode;
1563 1563 path = svn32->svn_path;
1564 1564
1565 1565 svn32->svn_timestamp = (uint32_t)svp->sv_timestamp;
1566 1566 svn32++;
1567 1567 } else {
1568 1568 nblocks = &svn->svn_nblocks;
1569 1569 mode = &svn->svn_mode;
1570 1570 path = svn->svn_path;
1571 1571
1572 1572 svn->svn_timestamp = svp->sv_timestamp;
1573 1573 svn++;
1574 1574 }
1575 1575
1576 1576 (void) strcpy(path, nsc_pathname(svp->sv_fd));
1577 1577 *nblocks = svp->sv_nblocks;
1578 1578 *mode = svp->sv_flag;
1579 1579
1580 1580 if (*nblocks == 0) {
1581 1581 if (sv_debug > 3)
1582 1582 cmn_err(CE_CONT, "!sv_list: need to reserve\n");
1583 1583
1584 1584 if (sv_reserve(svp->sv_fd, NSC_MULTI|NSC_PCATCH) == 0) {
1585 1585 *nblocks = svp->sv_nblocks;
1586 1586 nsc_release(svp->sv_fd);
1587 1587 }
1588 1588 }
1589 1589
1590 1590 if (++index >= size) {
1591 1591 /* Out of space */
1592 1592 (*extra)++;
1593 1593 }
1594 1594
1595 1595 rw_exit(&svp->sv_lock);
1596 1596 }
1597 1597 mutex_exit(&sv_mutex);
1598 1598
1599 1599 if (index < size) {
1600 1600 /* NULL terminated list */
1601 1601 if (ilp32)
1602 1602 svn32->svn_path[0] = '\0';
1603 1603 else
1604 1604 svn->svn_path[0] = '\0';
1605 1605 }
1606 1606
1607 1607 return (0);
1608 1608 }
1609 1609
1610 1610
1611 1611 static void
1612 1612 sv_thread_tune(int threads)
1613 1613 {
1614 1614 int incr = (threads > 0) ? 1 : -1;
1615 1615 int change = 0;
1616 1616 int nthreads;
1617 1617
1618 1618 ASSERT(MUTEX_HELD(&sv_mutex));
1619 1619
1620 1620 if (sv_threads_extra) {
1621 1621 /* keep track of any additional threads requested */
1622 1622 if (threads > 0) {
1623 1623 sv_threads_extra += threads;
1624 1624 return;
1625 1625 }
1626 1626 threads = -threads;
1627 1627 if (threads >= sv_threads_extra) {
1628 1628 threads -= sv_threads_extra;
1629 1629 sv_threads_extra = 0;
1630 1630 /* fall through to while loop */
1631 1631 } else {
1632 1632 sv_threads_extra -= threads;
1633 1633 return;
1634 1634 }
1635 1635 } else if (threads > 0) {
1636 1636 /*
1637 1637 * do not increase the number of threads beyond
1638 1638 * sv_threads_max when doing dynamic thread tuning
1639 1639 */
1640 1640 nthreads = nst_nthread(sv_tset);
1641 1641 if ((nthreads + threads) > sv_threads_max) {
1642 1642 sv_threads_extra = nthreads + threads - sv_threads_max;
1643 1643 threads = sv_threads_max - nthreads;
1644 1644 if (threads <= 0)
1645 1645 return;
1646 1646 }
1647 1647 }
1648 1648
1649 1649 if (threads < 0)
1650 1650 threads = -threads;
1651 1651
1652 1652 while (threads--) {
1653 1653 nthreads = nst_nthread(sv_tset);
1654 1654 sv_threads_needed += incr;
1655 1655
1656 1656 if (sv_threads_needed >= nthreads)
1657 1657 change += nst_add_thread(sv_tset, sv_threads_inc);
1658 1658 else if ((sv_threads_needed <
1659 1659 (nthreads - (sv_threads_inc + sv_threads_hysteresis))) &&
1660 1660 ((nthreads - sv_threads_inc) >= sv_threads))
1661 1661 change -= nst_del_thread(sv_tset, sv_threads_inc);
1662 1662 }
1663 1663
1664 1664 #ifdef DEBUG
1665 1665 if (change) {
1666 1666 cmn_err(CE_NOTE,
1667 1667 "!sv_thread_tune: threads needed %d, nthreads %d, "
1668 1668 "nthreads change %d",
1669 1669 sv_threads_needed, nst_nthread(sv_tset), change);
1670 1670 }
1671 1671 #endif
1672 1672 }
1673 1673
1674 1674
1675 1675 /* ARGSUSED */
1676 1676 static int
1677 1677 svopen(dev_t *devp, int flag, int otyp, cred_t *crp)
1678 1678 {
1679 1679 int rc;
1680 1680
1681 1681 mutex_enter(&sv_mutex);
1682 1682 rc = sv_init_devs();
1683 1683 mutex_exit(&sv_mutex);
1684 1684
1685 1685 return (rc);
1686 1686 }
1687 1687
1688 1688
1689 1689 /* ARGSUSED */
1690 1690 static int
1691 1691 svclose(dev_t dev, int flag, int otyp, cred_t *crp)
1692 1692 {
1693 1693 const int secs = HZ * 5;
1694 1694 const int ticks = HZ / 10;
1695 1695 int loops = secs / ticks;
1696 1696
1697 1697 mutex_enter(&sv_mutex);
1698 1698 while (sv_ndevices <= 0 && sv_tset != NULL && loops > 0) {
1699 1699 if (nst_nlive(sv_tset) <= 0) {
1700 1700 nst_destroy(sv_tset);
1701 1701 sv_tset = NULL;
1702 1702 break;
1703 1703 }
1704 1704
1705 1705 /* threads still active - wait for them to exit */
1706 1706 mutex_exit(&sv_mutex);
1707 1707 delay(ticks);
1708 1708 loops--;
1709 1709 mutex_enter(&sv_mutex);
1710 1710 }
1711 1711 mutex_exit(&sv_mutex);
1712 1712
1713 1713 if (loops <= 0) {
1714 1714 cmn_err(CE_WARN,
1715 1715 #ifndef DEBUG
1716 1716 /* do not write to console when non-DEBUG */
1717 1717 "!"
1718 1718 #endif
1719 1719 "sv:svclose: threads still active "
1720 1720 "after %d sec - leaking thread set", secs);
1721 1721 }
1722 1722
1723 1723 return (0);
1724 1724 }
1725 1725
1726 1726
1727 1727 static int
1728 1728 svioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *crp, int *rvalp)
1729 1729 {
1730 1730 char itmp1[12], itmp2[12]; /* temp char array for editing ints */
1731 1731 spcs_s_info_t kstatus; /* Kernel version of spcs status */
1732 1732 spcs_s_info_t ustatus; /* Address of user version of spcs status */
1733 1733 sv_list32_t svl32; /* 32 bit Initial structure for SVIOC_LIST */
1734 1734 sv_version_t svv; /* Version structure */
1735 1735 sv_conf_t svc; /* User config structure */
1736 1736 sv_list_t svl; /* Initial structure for SVIOC_LIST */
1737 1737 void *usvn; /* Address of user sv_name_t */
1738 1738 void *svn = NULL; /* Array for SVIOC_LIST */
1739 1739 uint64_t phash; /* pathname hash */
1740 1740 int rc = 0; /* Return code -- errno */
1741 1741 int size; /* Number of items in array */
1742 1742 int bytes; /* Byte size of array */
1743 1743 int ilp32; /* Convert data structures for ilp32 userland */
1744 1744
1745 1745 *rvalp = 0;
1746 1746
1747 1747 /*
1748 1748 * If sv_mod_status is 0 or SV_PREVENT_UNLOAD, then it will continue.
1749 1749 * else it means it previously was SV_PREVENT_UNLOAD, and now it's
1750 1750 * SV_ALLOW_UNLOAD, expecting the driver to eventually unload.
1751 1751 *
1752 1752 * SV_ALLOW_UNLOAD is final state, so no need to grab sv_mutex.
1753 1753 */
1754 1754 if (sv_mod_status == SV_ALLOW_UNLOAD) {
1755 1755 return (EBUSY);
1756 1756 }
1757 1757
1758 1758 if ((cmd != SVIOC_LIST) && ((rc = drv_priv(crp)) != 0))
1759 1759 return (rc);
1760 1760
1761 1761 kstatus = spcs_s_kcreate();
1762 1762 if (!kstatus) {
1763 1763 DTRACE_PROBE1(sv_ioctl_err_kcreate, dev_t, dev);
1764 1764 return (ENOMEM);
1765 1765 }
1766 1766
1767 1767 ilp32 = (ddi_model_convert_from((mode & FMODELS)) == DDI_MODEL_ILP32);
1768 1768
1769 1769 switch (cmd) {
1770 1770
1771 1771 case SVIOC_ENABLE:
1772 1772
1773 1773 if (ilp32) {
1774 1774 sv_conf32_t svc32;
1775 1775
1776 1776 if (ddi_copyin((void *)arg, &svc32,
1777 1777 sizeof (svc32), mode) < 0) {
1778 1778 spcs_s_kfree(kstatus);
1779 1779 return (EFAULT);
1780 1780 }
1781 1781
1782 1782 svc.svc_error = (spcs_s_info_t)svc32.svc_error;
1783 1783 (void) strcpy(svc.svc_path, svc32.svc_path);
1784 1784 svc.svc_flag = svc32.svc_flag;
1785 1785 svc.svc_major = svc32.svc_major;
1786 1786 svc.svc_minor = svc32.svc_minor;
1787 1787 } else {
1788 1788 if (ddi_copyin((void *)arg, &svc,
1789 1789 sizeof (svc), mode) < 0) {
1790 1790 spcs_s_kfree(kstatus);
1791 1791 return (EFAULT);
1792 1792 }
1793 1793 }
1794 1794
1795 1795 /* force to raw access */
1796 1796 svc.svc_flag = NSC_DEVICE;
1797 1797
1798 1798 if (sv_tset == NULL) {
1799 1799 mutex_enter(&sv_mutex);
1800 1800
1801 1801 if (sv_tset == NULL) {
1802 1802 sv_tset = nst_init("sv_thr", sv_threads);
1803 1803 }
1804 1804
1805 1805 mutex_exit(&sv_mutex);
1806 1806
1807 1807 if (sv_tset == NULL) {
1808 1808 cmn_err(CE_WARN,
1809 1809 "!sv: could not allocate %d threads",
1810 1810 sv_threads);
1811 1811 }
1812 1812 }
1813 1813
1814 1814 rc = sv_enable(svc.svc_path, svc.svc_flag,
1815 1815 makedevice(svc.svc_major, svc.svc_minor), kstatus);
1816 1816
1817 1817 if (rc == 0) {
1818 1818 sv_config_time = nsc_lbolt();
1819 1819
1820 1820 mutex_enter(&sv_mutex);
1821 1821 sv_thread_tune(sv_threads_dev);
1822 1822 mutex_exit(&sv_mutex);
1823 1823 }
1824 1824
1825 1825 DTRACE_PROBE3(sv_ioctl_end, dev_t, dev, int, *rvalp, int, rc);
1826 1826
1827 1827 return (spcs_s_ocopyoutf(&kstatus, svc.svc_error, rc));
1828 1828 /* NOTREACHED */
1829 1829
1830 1830 case SVIOC_DISABLE:
1831 1831
1832 1832 if (ilp32) {
1833 1833 sv_conf32_t svc32;
1834 1834
1835 1835 if (ddi_copyin((void *)arg, &svc32,
1836 1836 sizeof (svc32), mode) < 0) {
1837 1837 spcs_s_kfree(kstatus);
1838 1838 return (EFAULT);
1839 1839 }
1840 1840
1841 1841 svc.svc_error = (spcs_s_info_t)svc32.svc_error;
1842 1842 svc.svc_major = svc32.svc_major;
1843 1843 svc.svc_minor = svc32.svc_minor;
1844 1844 (void) strcpy(svc.svc_path, svc32.svc_path);
1845 1845 svc.svc_flag = svc32.svc_flag;
1846 1846 } else {
1847 1847 if (ddi_copyin((void *)arg, &svc,
1848 1848 sizeof (svc), mode) < 0) {
1849 1849 spcs_s_kfree(kstatus);
1850 1850 return (EFAULT);
1851 1851 }
1852 1852 }
1853 1853
1854 1854 if (svc.svc_major == (major_t)-1 &&
1855 1855 svc.svc_minor == (minor_t)-1) {
1856 1856 sv_dev_t *svp;
1857 1857 int i;
1858 1858
1859 1859 /*
1860 1860 * User level could not find the minor device
1861 1861 * node, so do this the slow way by searching
1862 1862 * the entire sv config for a matching pathname.
1863 1863 */
1864 1864
1865 1865 phash = nsc_strhash(svc.svc_path);
1866 1866
1867 1867 mutex_enter(&sv_mutex);
1868 1868
1869 1869 for (i = 0; i < sv_max_devices; i++) {
1870 1870 svp = &sv_devs[i];
1871 1871
1872 1872 if (svp->sv_state == SV_DISABLE ||
1873 1873 svp->sv_fd == NULL)
1874 1874 continue;
1875 1875
1876 1876 if (nsc_fdpathcmp(svp->sv_fd, phash,
1877 1877 svc.svc_path) == 0) {
1878 1878 svc.svc_major = getmajor(svp->sv_dev);
1879 1879 svc.svc_minor = getminor(svp->sv_dev);
1880 1880 break;
1881 1881 }
1882 1882 }
1883 1883
1884 1884 mutex_exit(&sv_mutex);
1885 1885
1886 1886 if (svc.svc_major == (major_t)-1 &&
1887 1887 svc.svc_minor == (minor_t)-1)
1888 1888 return (spcs_s_ocopyoutf(&kstatus,
1889 1889 svc.svc_error, SV_ENODEV));
1890 1890 }
1891 1891
1892 1892 rc = sv_disable(makedevice(svc.svc_major, svc.svc_minor),
1893 1893 kstatus);
1894 1894
1895 1895 if (rc == 0) {
1896 1896 sv_config_time = nsc_lbolt();
1897 1897
1898 1898 mutex_enter(&sv_mutex);
1899 1899 sv_thread_tune(-sv_threads_dev);
1900 1900 mutex_exit(&sv_mutex);
1901 1901 }
1902 1902
1903 1903 DTRACE_PROBE3(sv_ioctl_2, dev_t, dev, int, *rvalp, int, rc);
1904 1904
1905 1905 return (spcs_s_ocopyoutf(&kstatus, svc.svc_error, rc));
1906 1906 /* NOTREACHED */
1907 1907
1908 1908 case SVIOC_LIST:
1909 1909
1910 1910 if (ilp32) {
1911 1911 if (ddi_copyin((void *)arg, &svl32,
1912 1912 sizeof (svl32), mode) < 0) {
1913 1913 spcs_s_kfree(kstatus);
1914 1914 return (EFAULT);
1915 1915 }
1916 1916
1917 1917 ustatus = (spcs_s_info_t)svl32.svl_error;
1918 1918 size = svl32.svl_count;
1919 1919 usvn = (void *)(unsigned long)svl32.svl_names;
1920 1920 } else {
1921 1921 if (ddi_copyin((void *)arg, &svl,
1922 1922 sizeof (svl), mode) < 0) {
1923 1923 spcs_s_kfree(kstatus);
1924 1924 return (EFAULT);
1925 1925 }
1926 1926
1927 1927 ustatus = svl.svl_error;
1928 1928 size = svl.svl_count;
1929 1929 usvn = svl.svl_names;
1930 1930 }
1931 1931
1932 1932 /* Do some boundary checking */
1933 1933 if ((size < 0) || (size > sv_max_devices)) {
1934 1934 /* Array size is out of range */
1935 1935 return (spcs_s_ocopyoutf(&kstatus, ustatus,
1936 1936 SV_EARRBOUNDS, "0",
1937 1937 spcs_s_inttostring(sv_max_devices, itmp1,
1938 1938 sizeof (itmp1), 0),
1939 1939 spcs_s_inttostring(size, itmp2,
1940 1940 sizeof (itmp2), 0)));
↓ open down ↓ |
1940 lines elided |
↑ open up ↑ |
1941 1941 }
1942 1942
1943 1943 if (ilp32)
1944 1944 bytes = size * sizeof (sv_name32_t);
1945 1945 else
1946 1946 bytes = size * sizeof (sv_name_t);
1947 1947
1948 1948 /* Allocate memory for the array of structures */
1949 1949 if (bytes != 0) {
1950 1950 svn = kmem_zalloc(bytes, KM_SLEEP);
1951 - if (!svn) {
1952 - return (spcs_s_ocopyoutf(&kstatus,
1953 - ustatus, ENOMEM));
1954 - }
1955 1951 }
1956 1952
1957 1953 rc = sv_list(svn, size, rvalp, ilp32);
1958 1954 if (rc) {
1959 1955 if (svn != NULL)
1960 1956 kmem_free(svn, bytes);
1961 1957 return (spcs_s_ocopyoutf(&kstatus, ustatus, rc));
1962 1958 }
1963 1959
1964 1960 if (ilp32) {
1965 1961 svl32.svl_timestamp = (uint32_t)sv_config_time;
1966 1962 svl32.svl_maxdevs = (int32_t)sv_max_devices;
1967 1963
1968 1964 /* Return the list structure */
1969 1965 if (ddi_copyout(&svl32, (void *)arg,
1970 1966 sizeof (svl32), mode) < 0) {
1971 1967 spcs_s_kfree(kstatus);
1972 1968 if (svn != NULL)
1973 1969 kmem_free(svn, bytes);
1974 1970 return (EFAULT);
1975 1971 }
1976 1972 } else {
1977 1973 svl.svl_timestamp = sv_config_time;
1978 1974 svl.svl_maxdevs = sv_max_devices;
1979 1975
1980 1976 /* Return the list structure */
1981 1977 if (ddi_copyout(&svl, (void *)arg,
1982 1978 sizeof (svl), mode) < 0) {
1983 1979 spcs_s_kfree(kstatus);
1984 1980 if (svn != NULL)
1985 1981 kmem_free(svn, bytes);
1986 1982 return (EFAULT);
1987 1983 }
1988 1984 }
1989 1985
1990 1986 /* Return the array */
1991 1987 if (svn != NULL) {
1992 1988 if (ddi_copyout(svn, usvn, bytes, mode) < 0) {
1993 1989 kmem_free(svn, bytes);
1994 1990 spcs_s_kfree(kstatus);
1995 1991 return (EFAULT);
1996 1992 }
1997 1993 kmem_free(svn, bytes);
1998 1994 }
1999 1995
2000 1996 DTRACE_PROBE3(sv_ioctl_3, dev_t, dev, int, *rvalp, int, 0);
2001 1997
2002 1998 return (spcs_s_ocopyoutf(&kstatus, ustatus, 0));
2003 1999 /* NOTREACHED */
2004 2000
2005 2001 case SVIOC_VERSION:
2006 2002
2007 2003 if (ilp32) {
2008 2004 sv_version32_t svv32;
2009 2005
2010 2006 if (ddi_copyin((void *)arg, &svv32,
2011 2007 sizeof (svv32), mode) < 0) {
2012 2008 spcs_s_kfree(kstatus);
2013 2009 return (EFAULT);
2014 2010 }
2015 2011
2016 2012 svv32.svv_major_rev = sv_major_rev;
2017 2013 svv32.svv_minor_rev = sv_minor_rev;
2018 2014 svv32.svv_micro_rev = sv_micro_rev;
2019 2015 svv32.svv_baseline_rev = sv_baseline_rev;
2020 2016
2021 2017 if (ddi_copyout(&svv32, (void *)arg,
2022 2018 sizeof (svv32), mode) < 0) {
2023 2019 spcs_s_kfree(kstatus);
2024 2020 return (EFAULT);
2025 2021 }
2026 2022
2027 2023 ustatus = (spcs_s_info_t)svv32.svv_error;
2028 2024 } else {
2029 2025 if (ddi_copyin((void *)arg, &svv,
2030 2026 sizeof (svv), mode) < 0) {
2031 2027 spcs_s_kfree(kstatus);
2032 2028 return (EFAULT);
2033 2029 }
2034 2030
2035 2031 svv.svv_major_rev = sv_major_rev;
2036 2032 svv.svv_minor_rev = sv_minor_rev;
2037 2033 svv.svv_micro_rev = sv_micro_rev;
2038 2034 svv.svv_baseline_rev = sv_baseline_rev;
2039 2035
2040 2036 if (ddi_copyout(&svv, (void *)arg,
2041 2037 sizeof (svv), mode) < 0) {
2042 2038 spcs_s_kfree(kstatus);
2043 2039 return (EFAULT);
2044 2040 }
2045 2041
2046 2042 ustatus = svv.svv_error;
2047 2043 }
2048 2044
2049 2045 DTRACE_PROBE3(sv_ioctl_4, dev_t, dev, int, *rvalp, int, 0);
2050 2046
2051 2047 return (spcs_s_ocopyoutf(&kstatus, ustatus, 0));
2052 2048 /* NOTREACHED */
2053 2049
2054 2050 case SVIOC_UNLOAD:
2055 2051 rc = sv_prepare_unload();
2056 2052
2057 2053 if (ddi_copyout(&rc, (void *)arg, sizeof (rc), mode) < 0) {
2058 2054 rc = EFAULT;
2059 2055 }
2060 2056
2061 2057 spcs_s_kfree(kstatus);
2062 2058 return (rc);
2063 2059
2064 2060 default:
2065 2061 spcs_s_kfree(kstatus);
2066 2062
2067 2063 DTRACE_PROBE3(sv_ioctl_4, dev_t, dev, int, *rvalp, int, EINVAL);
2068 2064
2069 2065 return (EINVAL);
2070 2066 /* NOTREACHED */
2071 2067 }
2072 2068
2073 2069 /* NOTREACHED */
2074 2070 }
2075 2071
2076 2072
2077 2073 /* ARGSUSED */
2078 2074 static int
2079 2075 svprint(dev_t dev, char *str)
2080 2076 {
2081 2077 int instance = ddi_get_instance(sv_dip);
2082 2078 cmn_err(CE_WARN, "!%s%d: %s", ddi_get_name(sv_dip), instance, str);
2083 2079 return (0);
2084 2080 }
2085 2081
2086 2082
2087 2083 static void
2088 2084 _sv_lyr_strategy(struct buf *bp)
2089 2085 {
2090 2086 caddr_t buf_addr; /* pointer to linear buffer in bp */
2091 2087 nsc_buf_t *bufh = NULL;
2092 2088 nsc_buf_t *hndl = NULL;
2093 2089 sv_dev_t *svp;
2094 2090 nsc_vec_t *v;
2095 2091 sv_maj_t *maj;
2096 2092 nsc_size_t fba_req, fba_len; /* FBA lengths */
2097 2093 nsc_off_t fba_off; /* FBA offset */
2098 2094 size_t tocopy, nbytes; /* byte lengths */
2099 2095 int rw, rc; /* flags and return codes */
2100 2096 int (*fn)();
2101 2097
2102 2098 rc = 0;
2103 2099
2104 2100 if (sv_debug > 5)
2105 2101 cmn_err(CE_CONT, "!_sv_lyr_strategy(%p)\n", (void *)bp);
2106 2102
2107 2103 svp = sv_find_enabled(bp->b_edev, &maj);
2108 2104 if (svp == NULL) {
2109 2105 if (maj && (fn = maj->sm_strategy) != 0) {
2110 2106 if (!(maj->sm_flag & D_MP)) {
2111 2107 UNSAFE_ENTER();
2112 2108 rc = (*fn)(bp);
2113 2109 UNSAFE_EXIT();
2114 2110 } else {
2115 2111 rc = (*fn)(bp);
2116 2112 }
2117 2113 return;
2118 2114 } else {
2119 2115 bioerror(bp, ENODEV);
2120 2116 biodone(bp);
2121 2117 return;
2122 2118 }
2123 2119 }
2124 2120
2125 2121 ASSERT(RW_READ_HELD(&svp->sv_lock));
2126 2122
2127 2123 if (svp->sv_flag == 0) {
2128 2124 /*
2129 2125 * guard access mode
2130 2126 * - prevent user level access to the device
2131 2127 */
2132 2128 DTRACE_PROBE1(sv_lyr_strategy_err_guard, struct buf *, bp);
2133 2129 bioerror(bp, EPERM);
2134 2130 goto out;
2135 2131 }
2136 2132
2137 2133 if ((rc = sv_reserve(svp->sv_fd, NSC_MULTI|NSC_PCATCH)) != 0) {
2138 2134 DTRACE_PROBE1(sv_lyr_strategy_err_rsrv, struct buf *, bp);
2139 2135
2140 2136 if (rc == EINTR)
2141 2137 cmn_err(CE_WARN, "!nsc_reserve() returned EINTR");
2142 2138 bioerror(bp, rc);
2143 2139 goto out;
2144 2140 }
2145 2141
2146 2142 if (bp->b_lblkno >= (diskaddr_t)svp->sv_nblocks) {
2147 2143 DTRACE_PROBE1(sv_lyr_strategy_eof, struct buf *, bp);
2148 2144
2149 2145 if (bp->b_flags & B_READ) {
2150 2146 /* return EOF, not an error */
2151 2147 bp->b_resid = bp->b_bcount;
2152 2148 bioerror(bp, 0);
2153 2149 } else
2154 2150 bioerror(bp, EINVAL);
2155 2151
2156 2152 goto done;
2157 2153 }
2158 2154
2159 2155 /*
2160 2156 * Preallocate a handle once per call to strategy.
2161 2157 * If this fails, then the nsc_alloc_buf() will allocate
2162 2158 * a temporary handle per allocation/free pair.
2163 2159 */
2164 2160
2165 2161 DTRACE_PROBE1(sv_dbg_alloch_start, sv_dev_t *, svp);
2166 2162
2167 2163 bufh = nsc_alloc_handle(svp->sv_fd, NULL, NULL, NULL);
2168 2164
2169 2165 DTRACE_PROBE1(sv_dbg_alloch_end, sv_dev_t *, svp);
2170 2166
2171 2167 if (bufh && (bufh->sb_flag & NSC_HACTIVE) != 0) {
2172 2168 DTRACE_PROBE1(sv_lyr_strategy_err_hactive, struct buf *, bp);
2173 2169
2174 2170 cmn_err(CE_WARN,
2175 2171 "!sv: allocated active handle (bufh %p, flags %x)",
2176 2172 (void *)bufh, bufh->sb_flag);
2177 2173
2178 2174 bioerror(bp, ENXIO);
2179 2175 goto done;
2180 2176 }
2181 2177
2182 2178 fba_req = FBA_LEN(bp->b_bcount);
2183 2179 if (fba_req + bp->b_lblkno > (diskaddr_t)svp->sv_nblocks)
2184 2180 fba_req = (nsc_size_t)(svp->sv_nblocks - bp->b_lblkno);
2185 2181
2186 2182 rw = (bp->b_flags & B_READ) ? NSC_READ : NSC_WRITE;
2187 2183
2188 2184 bp_mapin(bp);
2189 2185
2190 2186 bp->b_resid = bp->b_bcount;
2191 2187 buf_addr = bp->b_un.b_addr;
2192 2188 fba_off = 0;
2193 2189
2194 2190 /*
2195 2191 * fba_req - requested size of transfer in FBAs after
2196 2192 * truncation to device extent, and allowing for
2197 2193 * possible non-FBA bounded final chunk.
2198 2194 * fba_off - offset of start of chunk from start of bp in FBAs.
2199 2195 * fba_len - size of this chunk in FBAs.
2200 2196 */
2201 2197
2202 2198 loop:
2203 2199 fba_len = min(fba_req, svp->sv_maxfbas);
2204 2200 hndl = bufh;
2205 2201
2206 2202 DTRACE_PROBE4(sv_dbg_allocb_start,
2207 2203 sv_dev_t *, svp,
2208 2204 uint64_t, (uint64_t)(bp->b_lblkno + fba_off),
2209 2205 uint64_t, (uint64_t)fba_len,
2210 2206 int, rw);
2211 2207
2212 2208 rc = nsc_alloc_buf(svp->sv_fd, (nsc_off_t)(bp->b_lblkno + fba_off),
2213 2209 fba_len, rw, &hndl);
2214 2210
2215 2211 DTRACE_PROBE1(sv_dbg_allocb_end, sv_dev_t *, svp);
2216 2212
2217 2213 if (rc > 0) {
2218 2214 DTRACE_PROBE1(sv_lyr_strategy_err_alloc, struct buf *, bp);
2219 2215 bioerror(bp, rc);
2220 2216 if (hndl != bufh)
2221 2217 (void) nsc_free_buf(hndl);
2222 2218 hndl = NULL;
2223 2219 goto done;
2224 2220 }
2225 2221
2226 2222 tocopy = min(FBA_SIZE(fba_len), bp->b_resid);
2227 2223 v = hndl->sb_vec;
2228 2224
2229 2225 if (rw == NSC_WRITE && FBA_OFF(tocopy) != 0) {
2230 2226 /*
2231 2227 * Not overwriting all of the last FBA, so read in the
2232 2228 * old contents now before we overwrite it with the new
2233 2229 * data.
2234 2230 */
2235 2231
2236 2232 DTRACE_PROBE2(sv_dbg_read_start, sv_dev_t *, svp,
2237 2233 uint64_t, (uint64_t)(hndl->sb_pos + hndl->sb_len - 1));
2238 2234
2239 2235 rc = nsc_read(hndl, (hndl->sb_pos + hndl->sb_len - 1), 1, 0);
2240 2236 if (rc > 0) {
2241 2237 bioerror(bp, rc);
2242 2238 goto done;
2243 2239 }
2244 2240
2245 2241 DTRACE_PROBE1(sv_dbg_read_end, sv_dev_t *, svp);
2246 2242 }
2247 2243
2248 2244 DTRACE_PROBE1(sv_dbg_bcopy_start, sv_dev_t *, svp);
2249 2245
2250 2246 while (tocopy > 0) {
2251 2247 nbytes = min(tocopy, (nsc_size_t)v->sv_len);
2252 2248
2253 2249 if (bp->b_flags & B_READ)
2254 2250 (void) bcopy(v->sv_addr, buf_addr, nbytes);
2255 2251 else
2256 2252 (void) bcopy(buf_addr, v->sv_addr, nbytes);
2257 2253
2258 2254 bp->b_resid -= nbytes;
2259 2255 buf_addr += nbytes;
2260 2256 tocopy -= nbytes;
2261 2257 v++;
2262 2258 }
2263 2259
2264 2260 DTRACE_PROBE1(sv_dbg_bcopy_end, sv_dev_t *, svp);
2265 2261
2266 2262 if ((bp->b_flags & B_READ) == 0) {
2267 2263 DTRACE_PROBE3(sv_dbg_write_start, sv_dev_t *, svp,
2268 2264 uint64_t, (uint64_t)hndl->sb_pos,
2269 2265 uint64_t, (uint64_t)hndl->sb_len);
2270 2266
2271 2267 rc = nsc_write(hndl, hndl->sb_pos, hndl->sb_len, 0);
2272 2268
2273 2269 DTRACE_PROBE1(sv_dbg_write_end, sv_dev_t *, svp);
2274 2270
2275 2271 if (rc > 0) {
2276 2272 bioerror(bp, rc);
2277 2273 goto done;
2278 2274 }
2279 2275 }
2280 2276
2281 2277 /*
2282 2278 * Adjust FBA offset and requested (ie. remaining) length,
2283 2279 * loop if more data to transfer.
2284 2280 */
2285 2281
2286 2282 fba_off += fba_len;
2287 2283 fba_req -= fba_len;
2288 2284
2289 2285 if (fba_req > 0) {
2290 2286 DTRACE_PROBE1(sv_dbg_freeb_start, sv_dev_t *, svp);
2291 2287
2292 2288 rc = nsc_free_buf(hndl);
2293 2289
2294 2290 DTRACE_PROBE1(sv_dbg_freeb_end, sv_dev_t *, svp);
2295 2291
2296 2292 if (rc > 0) {
2297 2293 DTRACE_PROBE1(sv_lyr_strategy_err_free,
2298 2294 struct buf *, bp);
2299 2295 bioerror(bp, rc);
2300 2296 }
2301 2297
2302 2298 hndl = NULL;
2303 2299
2304 2300 if (rc <= 0)
2305 2301 goto loop;
2306 2302 }
2307 2303
2308 2304 done:
2309 2305 if (hndl != NULL) {
2310 2306 DTRACE_PROBE1(sv_dbg_freeb_start, sv_dev_t *, svp);
2311 2307
2312 2308 rc = nsc_free_buf(hndl);
2313 2309
2314 2310 DTRACE_PROBE1(sv_dbg_freeb_end, sv_dev_t *, svp);
2315 2311
2316 2312 if (rc > 0) {
2317 2313 DTRACE_PROBE1(sv_lyr_strategy_err_free,
2318 2314 struct buf *, bp);
2319 2315 bioerror(bp, rc);
2320 2316 }
2321 2317
2322 2318 hndl = NULL;
2323 2319 }
2324 2320
2325 2321 if (bufh)
2326 2322 (void) nsc_free_handle(bufh);
2327 2323
2328 2324 DTRACE_PROBE1(sv_dbg_rlse_start, sv_dev_t *, svp);
2329 2325
2330 2326 nsc_release(svp->sv_fd);
2331 2327
2332 2328 DTRACE_PROBE1(sv_dbg_rlse_end, sv_dev_t *, svp);
2333 2329
2334 2330 out:
2335 2331 if (sv_debug > 5) {
2336 2332 cmn_err(CE_CONT,
2337 2333 "!_sv_lyr_strategy: bp %p, bufh %p, bp->b_error %d\n",
2338 2334 (void *)bp, (void *)bufh, bp->b_error);
2339 2335 }
2340 2336
2341 2337 DTRACE_PROBE2(sv_lyr_strategy_end, struct buf *, bp, int, bp->b_error);
2342 2338
2343 2339 rw_exit(&svp->sv_lock);
2344 2340 biodone(bp);
2345 2341 }
2346 2342
2347 2343
2348 2344 static void
2349 2345 sv_async_strategy(blind_t arg)
2350 2346 {
2351 2347 struct buf *bp = (struct buf *)arg;
2352 2348 _sv_lyr_strategy(bp);
2353 2349 }
2354 2350
2355 2351
2356 2352 static int
2357 2353 sv_lyr_strategy(struct buf *bp)
2358 2354 {
2359 2355 nsthread_t *tp;
2360 2356 int nlive;
2361 2357
2362 2358 /*
2363 2359 * If B_ASYNC was part of the DDI we could use it as a hint to
2364 2360 * not create a thread for synchronous i/o.
2365 2361 */
2366 2362 if (sv_dev_to_sv(bp->b_edev, NULL) == NULL) {
2367 2363 /* not sv enabled - just pass through */
2368 2364 DTRACE_PROBE1(sv_lyr_strategy_notsv, struct buf *, bp);
2369 2365 _sv_lyr_strategy(bp);
2370 2366 return (0);
2371 2367 }
2372 2368
2373 2369 if (sv_debug > 4) {
2374 2370 cmn_err(CE_CONT, "!sv_lyr_strategy: nthread %d nlive %d\n",
2375 2371 nst_nthread(sv_tset), nst_nlive(sv_tset));
2376 2372 }
2377 2373
2378 2374 /*
2379 2375 * If there are only guard devices enabled there
2380 2376 * won't be a threadset, so don't try and use it.
2381 2377 */
2382 2378 tp = NULL;
2383 2379 if (sv_tset != NULL) {
2384 2380 tp = nst_create(sv_tset, sv_async_strategy, (blind_t)bp, 0);
2385 2381 }
2386 2382
2387 2383 if (tp == NULL) {
2388 2384 /*
2389 2385 * out of threads, so fall back to synchronous io.
2390 2386 */
2391 2387 if (sv_debug > 0) {
2392 2388 cmn_err(CE_CONT,
2393 2389 "!sv_lyr_strategy: thread alloc failed\n");
2394 2390 }
2395 2391
2396 2392 DTRACE_PROBE1(sv_lyr_strategy_no_thread,
2397 2393 struct buf *, bp);
2398 2394
2399 2395 _sv_lyr_strategy(bp);
2400 2396 sv_no_threads++;
2401 2397 } else {
2402 2398 nlive = nst_nlive(sv_tset);
2403 2399 if (nlive > sv_max_nlive) {
2404 2400 if (sv_debug > 0) {
2405 2401 cmn_err(CE_CONT,
2406 2402 "!sv_lyr_strategy: "
2407 2403 "new max nlive %d (nthread %d)\n",
2408 2404 nlive, nst_nthread(sv_tset));
2409 2405 }
2410 2406
2411 2407 sv_max_nlive = nlive;
2412 2408 }
2413 2409 }
2414 2410
2415 2411 return (0);
2416 2412 }
2417 2413
2418 2414
2419 2415 #ifndef offsetof
2420 2416 #define offsetof(s, m) ((size_t)(&((s *)0)->m))
2421 2417 #endif
2422 2418
2423 2419 /*
2424 2420 * re-write the size of the current partition
2425 2421 */
2426 2422 static int
2427 2423 sv_fix_dkiocgvtoc(const intptr_t arg, const int mode, sv_dev_t *svp)
2428 2424 {
2429 2425 size_t offset;
2430 2426 int ilp32;
2431 2427 int pnum;
2432 2428 int rc;
2433 2429
2434 2430 ilp32 = (ddi_model_convert_from((mode & FMODELS)) == DDI_MODEL_ILP32);
2435 2431
2436 2432 rc = nskern_partition(svp->sv_dev, &pnum);
2437 2433 if (rc != 0) {
2438 2434 return (rc);
2439 2435 }
2440 2436
2441 2437 if (pnum < 0 || pnum >= V_NUMPAR) {
2442 2438 cmn_err(CE_WARN,
2443 2439 "!sv_gvtoc: unable to determine partition number "
2444 2440 "for dev %lx", svp->sv_dev);
2445 2441 return (EINVAL);
2446 2442 }
2447 2443
2448 2444 if (ilp32) {
2449 2445 int32_t p_size;
2450 2446
2451 2447 #ifdef _SunOS_5_6
2452 2448 offset = offsetof(struct vtoc, v_part);
2453 2449 offset += sizeof (struct partition) * pnum;
2454 2450 offset += offsetof(struct partition, p_size);
2455 2451 #else
2456 2452 offset = offsetof(struct vtoc32, v_part);
2457 2453 offset += sizeof (struct partition32) * pnum;
2458 2454 offset += offsetof(struct partition32, p_size);
2459 2455 #endif
2460 2456
2461 2457 p_size = (int32_t)svp->sv_nblocks;
2462 2458 if (p_size == 0) {
2463 2459 if (sv_reserve(svp->sv_fd,
2464 2460 NSC_MULTI|NSC_PCATCH) == 0) {
2465 2461 p_size = (int32_t)svp->sv_nblocks;
2466 2462 nsc_release(svp->sv_fd);
2467 2463 } else {
2468 2464 rc = EINTR;
2469 2465 }
2470 2466 }
2471 2467
2472 2468 if ((rc == 0) && ddi_copyout(&p_size, (void *)(arg + offset),
2473 2469 sizeof (p_size), mode) != 0) {
2474 2470 rc = EFAULT;
2475 2471 }
2476 2472 } else {
2477 2473 long p_size;
2478 2474
2479 2475 offset = offsetof(struct vtoc, v_part);
2480 2476 offset += sizeof (struct partition) * pnum;
2481 2477 offset += offsetof(struct partition, p_size);
2482 2478
2483 2479 p_size = (long)svp->sv_nblocks;
2484 2480 if (p_size == 0) {
2485 2481 if (sv_reserve(svp->sv_fd,
2486 2482 NSC_MULTI|NSC_PCATCH) == 0) {
2487 2483 p_size = (long)svp->sv_nblocks;
2488 2484 nsc_release(svp->sv_fd);
2489 2485 } else {
2490 2486 rc = EINTR;
2491 2487 }
2492 2488 }
2493 2489
2494 2490 if ((rc == 0) && ddi_copyout(&p_size, (void *)(arg + offset),
2495 2491 sizeof (p_size), mode) != 0) {
2496 2492 rc = EFAULT;
2497 2493 }
2498 2494 }
2499 2495
2500 2496 return (rc);
2501 2497 }
2502 2498
2503 2499
2504 2500 #ifdef DKIOCPARTITION
2505 2501 /*
2506 2502 * re-write the size of the current partition
2507 2503 *
2508 2504 * arg is dk_efi_t.
2509 2505 *
2510 2506 * dk_efi_t->dki_data = (void *)(uintptr_t)efi.dki_data_64;
2511 2507 *
2512 2508 * dk_efi_t->dki_data --> efi_gpt_t (label header)
2513 2509 * dk_efi_t->dki_data + 1 --> efi_gpe_t[] (array of partitions)
2514 2510 *
2515 2511 * efi_gpt_t->efi_gpt_PartitionEntryArrayCRC32 --> CRC32 of array of parts
2516 2512 * efi_gpt_t->efi_gpt_HeaderCRC32 --> CRC32 of header itself
2517 2513 *
2518 2514 * This assumes that sizeof (efi_gpt_t) is the same as the size of a
2519 2515 * logical block on the disk.
2520 2516 *
2521 2517 * Everything is little endian (i.e. disk format).
2522 2518 */
2523 2519 static int
2524 2520 sv_fix_dkiocgetefi(const intptr_t arg, const int mode, sv_dev_t *svp)
2525 2521 {
2526 2522 dk_efi_t efi;
2527 2523 efi_gpt_t gpt;
2528 2524 efi_gpe_t *gpe = NULL;
2529 2525 size_t sgpe;
2530 2526 uint64_t p_size; /* virtual partition size from nsctl */
2531 2527 uint32_t crc;
2532 2528 int unparts; /* number of parts in user's array */
2533 2529 int pnum;
2534 2530 int rc;
2535 2531
2536 2532 rc = nskern_partition(svp->sv_dev, &pnum);
2537 2533 if (rc != 0) {
2538 2534 return (rc);
2539 2535 }
2540 2536
2541 2537 if (pnum < 0) {
2542 2538 cmn_err(CE_WARN,
2543 2539 "!sv_efi: unable to determine partition number for dev %lx",
2544 2540 svp->sv_dev);
2545 2541 return (EINVAL);
2546 2542 }
2547 2543
2548 2544 if (ddi_copyin((void *)arg, &efi, sizeof (efi), mode)) {
2549 2545 return (EFAULT);
2550 2546 }
2551 2547
2552 2548 efi.dki_data = (void *)(uintptr_t)efi.dki_data_64;
2553 2549
2554 2550 if (efi.dki_length < sizeof (gpt) + sizeof (gpe)) {
2555 2551 return (EINVAL);
2556 2552 }
2557 2553
2558 2554 if (ddi_copyin((void *)efi.dki_data, &gpt, sizeof (gpt), mode)) {
2559 2555 rc = EFAULT;
2560 2556 goto out;
2561 2557 }
2562 2558
2563 2559 if ((unparts = LE_32(gpt.efi_gpt_NumberOfPartitionEntries)) == 0)
2564 2560 unparts = 1;
2565 2561 else if (pnum >= unparts) {
2566 2562 cmn_err(CE_WARN,
2567 2563 "!sv_efi: partition# beyond end of user array (%d >= %d)",
2568 2564 pnum, unparts);
2569 2565 return (EINVAL);
2570 2566 }
2571 2567
2572 2568 sgpe = sizeof (*gpe) * unparts;
2573 2569 gpe = kmem_alloc(sgpe, KM_SLEEP);
2574 2570
2575 2571 if (ddi_copyin((void *)(efi.dki_data + 1), gpe, sgpe, mode)) {
2576 2572 rc = EFAULT;
2577 2573 goto out;
2578 2574 }
2579 2575
2580 2576 p_size = svp->sv_nblocks;
2581 2577 if (p_size == 0) {
2582 2578 if (sv_reserve(svp->sv_fd, NSC_MULTI|NSC_PCATCH) == 0) {
2583 2579 p_size = (diskaddr_t)svp->sv_nblocks;
2584 2580 nsc_release(svp->sv_fd);
2585 2581 } else {
2586 2582 rc = EINTR;
2587 2583 }
2588 2584 }
2589 2585
2590 2586 gpe[pnum].efi_gpe_EndingLBA = LE_64(
2591 2587 LE_64(gpe[pnum].efi_gpe_StartingLBA) + p_size - 1);
2592 2588
2593 2589 gpt.efi_gpt_PartitionEntryArrayCRC32 = 0;
2594 2590 CRC32(crc, gpe, sgpe, -1U, sv_crc32_table);
2595 2591 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
2596 2592
2597 2593 gpt.efi_gpt_HeaderCRC32 = 0;
2598 2594 CRC32(crc, &gpt, sizeof (gpt), -1U, sv_crc32_table);
2599 2595 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
2600 2596
2601 2597 if ((rc == 0) && ddi_copyout(&gpt, efi.dki_data, sizeof (gpt), mode)) {
2602 2598 rc = EFAULT;
2603 2599 goto out;
2604 2600 }
2605 2601
2606 2602 if ((rc == 0) && ddi_copyout(gpe, efi.dki_data + 1, sgpe, mode)) {
2607 2603 rc = EFAULT;
2608 2604 goto out;
2609 2605 }
2610 2606
2611 2607 out:
2612 2608 if (gpe) {
2613 2609 kmem_free(gpe, sgpe);
2614 2610 }
2615 2611
2616 2612 return (rc);
2617 2613 }
2618 2614
2619 2615
2620 2616 /*
2621 2617 * Re-write the size of the partition specified by p_partno
2622 2618 *
2623 2619 * Note that if a DKIOCPARTITION is issued to an fd opened against a
2624 2620 * non-sv'd device, but p_partno requests the size for a different
2625 2621 * device that is sv'd, this function will *not* be called as sv is
2626 2622 * not interposed on the original device (the fd).
2627 2623 *
2628 2624 * It would not be easy to change this as we cannot get the partition
2629 2625 * number for the non-sv'd device, so cannot compute the dev_t of the
2630 2626 * (sv'd) p_partno device, and so cannot find out if it is sv'd or get
2631 2627 * its size from nsctl.
2632 2628 *
2633 2629 * See also the "Bug 4755783" comment in sv_lyr_ioctl().
2634 2630 */
2635 2631 static int
2636 2632 sv_fix_dkiocpartition(const intptr_t arg, const int mode, sv_dev_t *svp)
2637 2633 {
2638 2634 struct partition64 p64;
2639 2635 sv_dev_t *nsvp = NULL;
2640 2636 diskaddr_t p_size;
2641 2637 minor_t nminor;
2642 2638 int pnum, rc;
2643 2639 dev_t ndev;
2644 2640
2645 2641 rc = nskern_partition(svp->sv_dev, &pnum);
2646 2642 if (rc != 0) {
2647 2643 return (rc);
2648 2644 }
2649 2645
2650 2646 if (ddi_copyin((void *)arg, &p64, sizeof (p64), mode)) {
2651 2647 return (EFAULT);
2652 2648 }
2653 2649
2654 2650 if (p64.p_partno != pnum) {
2655 2651 /* switch to requested partition, not the current one */
2656 2652 nminor = getminor(svp->sv_dev) + (p64.p_partno - pnum);
2657 2653 ndev = makedevice(getmajor(svp->sv_dev), nminor);
2658 2654 nsvp = sv_find_enabled(ndev, NULL);
2659 2655 if (nsvp == NULL) {
2660 2656 /* not sv device - just return */
2661 2657 return (0);
2662 2658 }
2663 2659
2664 2660 svp = nsvp;
2665 2661 }
2666 2662
2667 2663 p_size = svp->sv_nblocks;
2668 2664 if (p_size == 0) {
2669 2665 if (sv_reserve(svp->sv_fd, NSC_MULTI|NSC_PCATCH) == 0) {
2670 2666 p_size = (diskaddr_t)svp->sv_nblocks;
2671 2667 nsc_release(svp->sv_fd);
2672 2668 } else {
2673 2669 rc = EINTR;
2674 2670 }
2675 2671 }
2676 2672
2677 2673 if (nsvp != NULL) {
2678 2674 rw_exit(&nsvp->sv_lock);
2679 2675 }
2680 2676
2681 2677 if ((rc == 0) && ddi_copyout(&p_size,
2682 2678 (void *)(arg + offsetof(struct partition64, p_size)),
2683 2679 sizeof (p_size), mode) != 0) {
2684 2680 return (EFAULT);
2685 2681 }
2686 2682
2687 2683 return (rc);
2688 2684 }
2689 2685 #endif /* DKIOCPARTITION */
2690 2686
2691 2687
2692 2688 static int
2693 2689 sv_lyr_ioctl(const dev_t dev, const int cmd, const intptr_t arg,
2694 2690 const int mode, cred_t *crp, int *rvalp)
2695 2691 {
2696 2692 sv_dev_t *svp;
2697 2693 sv_maj_t *maj;
2698 2694 int (*fn)();
2699 2695 int rc = 0;
2700 2696
2701 2697 maj = 0;
2702 2698 fn = 0;
2703 2699
2704 2700 /*
2705 2701 * If sv_mod_status is 0 or SV_PREVENT_UNLOAD, then it will continue.
2706 2702 * else it means it previously was SV_PREVENT_UNLOAD, and now it's
2707 2703 * SV_ALLOW_UNLOAD, expecting the driver to eventually unload.
2708 2704 *
2709 2705 * SV_ALLOW_UNLOAD is final state, so no need to grab sv_mutex.
2710 2706 */
2711 2707 if (sv_mod_status == SV_ALLOW_UNLOAD) {
2712 2708 return (EBUSY);
2713 2709 }
2714 2710
2715 2711 svp = sv_find_enabled(dev, &maj);
2716 2712 if (svp != NULL) {
2717 2713 if (nskernd_isdaemon()) {
2718 2714 /*
2719 2715 * This is nskernd which always needs to see
2720 2716 * the underlying disk device accurately.
2721 2717 *
2722 2718 * So just pass the ioctl straight through
2723 2719 * to the underlying driver as though the device
2724 2720 * was not sv enabled.
2725 2721 */
2726 2722 DTRACE_PROBE2(sv_lyr_ioctl_nskernd, sv_dev_t *, svp,
2727 2723 dev_t, dev);
2728 2724
2729 2725 rw_exit(&svp->sv_lock);
2730 2726 svp = NULL;
2731 2727 } else {
2732 2728 ASSERT(RW_READ_HELD(&svp->sv_lock));
2733 2729 }
2734 2730 }
2735 2731
2736 2732 /*
2737 2733 * We now have a locked and enabled SV device, or a non-SV device.
2738 2734 */
2739 2735
2740 2736 switch (cmd) {
2741 2737 /*
2742 2738 * DKIOCGVTOC, DKIOCSVTOC, DKIOCPARTITION, DKIOCGETEFI
2743 2739 * and DKIOCSETEFI are intercepted and faked up as some
2744 2740 * i/o providers emulate volumes of a different size to
2745 2741 * the underlying volume.
2746 2742 *
2747 2743 * Setting the size by rewriting the vtoc is not permitted.
2748 2744 */
2749 2745
2750 2746 case DKIOCSVTOC:
2751 2747 #ifdef DKIOCPARTITION
2752 2748 case DKIOCSETEFI:
2753 2749 #endif
2754 2750 if (svp == NULL) {
2755 2751 /* not intercepted -- allow ioctl through */
2756 2752 break;
2757 2753 }
2758 2754
2759 2755 rw_exit(&svp->sv_lock);
2760 2756
2761 2757 DTRACE_PROBE2(sv_lyr_ioctl_svtoc, dev_t, dev, int, EPERM);
2762 2758
2763 2759 return (EPERM);
2764 2760
2765 2761 default:
2766 2762 break;
2767 2763 }
2768 2764
2769 2765 /*
2770 2766 * Pass through the real ioctl command.
2771 2767 */
2772 2768
2773 2769 if (maj && (fn = maj->sm_ioctl) != 0) {
2774 2770 if (!(maj->sm_flag & D_MP)) {
2775 2771 UNSAFE_ENTER();
2776 2772 rc = (*fn)(dev, cmd, arg, mode, crp, rvalp);
2777 2773 UNSAFE_EXIT();
2778 2774 } else {
2779 2775 rc = (*fn)(dev, cmd, arg, mode, crp, rvalp);
2780 2776 }
2781 2777 } else {
2782 2778 rc = ENODEV;
2783 2779 }
2784 2780
2785 2781 /*
2786 2782 * Bug 4755783
2787 2783 * Fix up the size of the current partition to allow
2788 2784 * for the virtual volume to be a different size to the
2789 2785 * physical volume (e.g. for II compact dependent shadows).
2790 2786 *
2791 2787 * Note that this only attempts to fix up the current partition
2792 2788 * - the one that the ioctl was issued against. There could be
2793 2789 * other sv'd partitions in the same vtoc, but we cannot tell
2794 2790 * so we don't attempt to fix them up.
2795 2791 */
2796 2792
2797 2793 if (svp != NULL && rc == 0) {
2798 2794 switch (cmd) {
2799 2795 case DKIOCGVTOC:
2800 2796 rc = sv_fix_dkiocgvtoc(arg, mode, svp);
2801 2797 break;
2802 2798
2803 2799 #ifdef DKIOCPARTITION
2804 2800 case DKIOCGETEFI:
2805 2801 rc = sv_fix_dkiocgetefi(arg, mode, svp);
2806 2802 break;
2807 2803
2808 2804 case DKIOCPARTITION:
2809 2805 rc = sv_fix_dkiocpartition(arg, mode, svp);
2810 2806 break;
2811 2807 #endif /* DKIOCPARTITION */
2812 2808 }
2813 2809 }
2814 2810
2815 2811 if (svp != NULL) {
2816 2812 rw_exit(&svp->sv_lock);
2817 2813 }
2818 2814
2819 2815 return (rc);
2820 2816 }
↓ open down ↓ |
856 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX