xref: /freebsd/sys/dev/nvme/nvme_ns.c (revision a8675d927bbde29d5a4dc6efddf2f0dc6d6d6983) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2012-2013 Intel Corporation
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/bio.h>
34 #include <sys/bus.h>
35 #include <sys/conf.h>
36 #include <sys/disk.h>
37 #include <sys/fcntl.h>
38 #include <sys/ioccom.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
41 #include <sys/proc.h>
42 #include <sys/systm.h>
43 
44 #include <dev/pci/pcivar.h>
45 
46 #include <geom/geom.h>
47 
48 #include "nvme_private.h"
49 
50 static void		nvme_bio_child_inbed(struct bio *parent, int bio_error);
51 static void		nvme_bio_child_done(void *arg,
52 					    const struct nvme_completion *cpl);
53 static uint32_t		nvme_get_num_segments(uint64_t addr, uint64_t size,
54 					      uint32_t alignment);
55 static void		nvme_free_child_bios(int num_bios,
56 					     struct bio **child_bios);
57 static struct bio **	nvme_allocate_child_bios(int num_bios);
58 static struct bio **	nvme_construct_child_bios(struct bio *bp,
59 						  uint32_t alignment,
60 						  int *num_bios);
61 static int		nvme_ns_split_bio(struct nvme_namespace *ns,
62 					  struct bio *bp,
63 					  uint32_t alignment);
64 
65 static int
66 nvme_ns_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag,
67     struct thread *td)
68 {
69 	struct nvme_namespace			*ns;
70 	struct nvme_controller			*ctrlr;
71 	struct nvme_pt_command			*pt;
72 
73 	ns = cdev->si_drv1;
74 	ctrlr = ns->ctrlr;
75 
76 	switch (cmd) {
77 	case NVME_IO_TEST:
78 	case NVME_BIO_TEST:
79 		nvme_ns_test(ns, cmd, arg);
80 		break;
81 	case NVME_PASSTHROUGH_CMD:
82 		pt = (struct nvme_pt_command *)arg;
83 		return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, ns->id,
84 		    1 /* is_user_buffer */, 0 /* is_admin_cmd */));
85 	case NVME_GET_NSID:
86 	{
87 		struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)arg;
88 		strncpy(gnsid->cdev, device_get_nameunit(ctrlr->dev),
89 		    sizeof(gnsid->cdev));
90 		gnsid->nsid = ns->id;
91 		break;
92 	}
93 	case DIOCGMEDIASIZE:
94 		*(off_t *)arg = (off_t)nvme_ns_get_size(ns);
95 		break;
96 	case DIOCGSECTORSIZE:
97 		*(u_int *)arg = nvme_ns_get_sector_size(ns);
98 		break;
99 	default:
100 		return (ENOTTY);
101 	}
102 
103 	return (0);
104 }
105 
106 static int
107 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused,
108     struct thread *td)
109 {
110 	int error = 0;
111 
112 	if (flags & FWRITE)
113 		error = securelevel_gt(td->td_ucred, 0);
114 
115 	return (error);
116 }
117 
118 static int
119 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused,
120     struct thread *td)
121 {
122 
123 	return (0);
124 }
125 
126 static void
127 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl)
128 {
129 	struct bio *bp = arg;
130 
131 	/*
132 	 * TODO: add more extensive translation of NVMe status codes
133 	 *  to different bio error codes (i.e. EIO, EINVAL, etc.)
134 	 */
135 	if (nvme_completion_is_error(cpl)) {
136 		bp->bio_error = EIO;
137 		bp->bio_flags |= BIO_ERROR;
138 		bp->bio_resid = bp->bio_bcount;
139 	} else
140 		bp->bio_resid = 0;
141 
142 	biodone(bp);
143 }
144 
145 static void
146 nvme_ns_strategy(struct bio *bp)
147 {
148 	struct nvme_namespace	*ns;
149 	int			err;
150 
151 	ns = bp->bio_dev->si_drv1;
152 	err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done);
153 
154 	if (err) {
155 		bp->bio_error = err;
156 		bp->bio_flags |= BIO_ERROR;
157 		bp->bio_resid = bp->bio_bcount;
158 		biodone(bp);
159 	}
160 
161 }
162 
163 static struct cdevsw nvme_ns_cdevsw = {
164 	.d_version =	D_VERSION,
165 	.d_flags =	D_DISK,
166 	.d_read =	physread,
167 	.d_write =	physwrite,
168 	.d_open =	nvme_ns_open,
169 	.d_close =	nvme_ns_close,
170 	.d_strategy =	nvme_ns_strategy,
171 	.d_ioctl =	nvme_ns_ioctl
172 };
173 
174 uint32_t
175 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns)
176 {
177 	return ns->ctrlr->max_xfer_size;
178 }
179 
180 uint32_t
181 nvme_ns_get_sector_size(struct nvme_namespace *ns)
182 {
183 	uint8_t flbas_fmt, lbads;
184 
185 	flbas_fmt = (ns->data.flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
186 		NVME_NS_DATA_FLBAS_FORMAT_MASK;
187 	lbads = (ns->data.lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) &
188 		NVME_NS_DATA_LBAF_LBADS_MASK;
189 
190 	return (1 << lbads);
191 }
192 
193 uint64_t
194 nvme_ns_get_num_sectors(struct nvme_namespace *ns)
195 {
196 	return (ns->data.nsze);
197 }
198 
199 uint64_t
200 nvme_ns_get_size(struct nvme_namespace *ns)
201 {
202 	return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns));
203 }
204 
205 uint32_t
206 nvme_ns_get_flags(struct nvme_namespace *ns)
207 {
208 	return (ns->flags);
209 }
210 
211 const char *
212 nvme_ns_get_serial_number(struct nvme_namespace *ns)
213 {
214 	return ((const char *)ns->ctrlr->cdata.sn);
215 }
216 
217 const char *
218 nvme_ns_get_model_number(struct nvme_namespace *ns)
219 {
220 	return ((const char *)ns->ctrlr->cdata.mn);
221 }
222 
223 const struct nvme_namespace_data *
224 nvme_ns_get_data(struct nvme_namespace *ns)
225 {
226 
227 	return (&ns->data);
228 }
229 
230 uint32_t
231 nvme_ns_get_stripesize(struct nvme_namespace *ns)
232 {
233 
234 	if (((ns->data.nsfeat >> NVME_NS_DATA_NSFEAT_NPVALID_SHIFT) &
235 	    NVME_NS_DATA_NSFEAT_NPVALID_MASK) != 0 && ns->data.npwg != 0) {
236 		return ((ns->data.npwg + 1) * nvme_ns_get_sector_size(ns));
237 	}
238 	return (ns->boundary);
239 }
240 
241 static void
242 nvme_ns_bio_done(void *arg, const struct nvme_completion *status)
243 {
244 	struct bio	*bp = arg;
245 	nvme_cb_fn_t	bp_cb_fn;
246 
247 	bp_cb_fn = bp->bio_driver1;
248 
249 	if (bp->bio_driver2)
250 		free(bp->bio_driver2, M_NVME);
251 
252 	if (nvme_completion_is_error(status)) {
253 		bp->bio_flags |= BIO_ERROR;
254 		if (bp->bio_error == 0)
255 			bp->bio_error = EIO;
256 	}
257 
258 	if ((bp->bio_flags & BIO_ERROR) == 0)
259 		bp->bio_resid = 0;
260 	else
261 		bp->bio_resid = bp->bio_bcount;
262 
263 	bp_cb_fn(bp, status);
264 }
265 
266 static void
267 nvme_bio_child_inbed(struct bio *parent, int bio_error)
268 {
269 	struct nvme_completion	parent_cpl;
270 	int			children, inbed;
271 
272 	if (bio_error != 0) {
273 		parent->bio_flags |= BIO_ERROR;
274 		parent->bio_error = bio_error;
275 	}
276 
277 	/*
278 	 * atomic_fetchadd will return value before adding 1, so we still
279 	 *  must add 1 to get the updated inbed number.  Save bio_children
280 	 *  before incrementing to guard against race conditions when
281 	 *  two children bios complete on different queues.
282 	 */
283 	children = atomic_load_acq_int(&parent->bio_children);
284 	inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1;
285 	if (inbed == children) {
286 		bzero(&parent_cpl, sizeof(parent_cpl));
287 		if (parent->bio_flags & BIO_ERROR) {
288 			parent_cpl.status &= ~(NVME_STATUS_SC_MASK << NVME_STATUS_SC_SHIFT);
289 			parent_cpl.status |= (NVME_SC_DATA_TRANSFER_ERROR) << NVME_STATUS_SC_SHIFT;
290 		}
291 		nvme_ns_bio_done(parent, &parent_cpl);
292 	}
293 }
294 
295 static void
296 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl)
297 {
298 	struct bio		*child = arg;
299 	struct bio		*parent;
300 	int			bio_error;
301 
302 	parent = child->bio_parent;
303 	g_destroy_bio(child);
304 	bio_error = nvme_completion_is_error(cpl) ? EIO : 0;
305 	nvme_bio_child_inbed(parent, bio_error);
306 }
307 
308 static uint32_t
309 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align)
310 {
311 	uint32_t	num_segs, offset, remainder;
312 
313 	if (align == 0)
314 		return (1);
315 
316 	KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n"));
317 
318 	num_segs = size / align;
319 	remainder = size & (align - 1);
320 	offset = addr & (align - 1);
321 	if (remainder > 0 || offset > 0)
322 		num_segs += 1 + (remainder + offset - 1) / align;
323 	return (num_segs);
324 }
325 
326 static void
327 nvme_free_child_bios(int num_bios, struct bio **child_bios)
328 {
329 	int i;
330 
331 	for (i = 0; i < num_bios; i++) {
332 		if (child_bios[i] != NULL)
333 			g_destroy_bio(child_bios[i]);
334 	}
335 
336 	free(child_bios, M_NVME);
337 }
338 
339 static struct bio **
340 nvme_allocate_child_bios(int num_bios)
341 {
342 	struct bio **child_bios;
343 	int err = 0, i;
344 
345 	child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT);
346 	if (child_bios == NULL)
347 		return (NULL);
348 
349 	for (i = 0; i < num_bios; i++) {
350 		child_bios[i] = g_new_bio();
351 		if (child_bios[i] == NULL)
352 			err = ENOMEM;
353 	}
354 
355 	if (err == ENOMEM) {
356 		nvme_free_child_bios(num_bios, child_bios);
357 		return (NULL);
358 	}
359 
360 	return (child_bios);
361 }
362 
363 static struct bio **
364 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios)
365 {
366 	struct bio	**child_bios;
367 	struct bio	*child;
368 	uint64_t	cur_offset;
369 	caddr_t		data;
370 	uint32_t	rem_bcount;
371 	int		i;
372 	struct vm_page	**ma;
373 	uint32_t	ma_offset;
374 
375 	*num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount,
376 	    alignment);
377 	child_bios = nvme_allocate_child_bios(*num_bios);
378 	if (child_bios == NULL)
379 		return (NULL);
380 
381 	bp->bio_children = *num_bios;
382 	bp->bio_inbed = 0;
383 	cur_offset = bp->bio_offset;
384 	rem_bcount = bp->bio_bcount;
385 	data = bp->bio_data;
386 	ma_offset = bp->bio_ma_offset;
387 	ma = bp->bio_ma;
388 
389 	for (i = 0; i < *num_bios; i++) {
390 		child = child_bios[i];
391 		child->bio_parent = bp;
392 		child->bio_cmd = bp->bio_cmd;
393 		child->bio_offset = cur_offset;
394 		child->bio_bcount = min(rem_bcount,
395 		    alignment - (cur_offset & (alignment - 1)));
396 		child->bio_flags = bp->bio_flags;
397 		if (bp->bio_flags & BIO_UNMAPPED) {
398 			child->bio_ma_offset = ma_offset;
399 			child->bio_ma = ma;
400 			child->bio_ma_n =
401 			    nvme_get_num_segments(child->bio_ma_offset,
402 				child->bio_bcount, PAGE_SIZE);
403 			ma_offset = (ma_offset + child->bio_bcount) &
404 			    PAGE_MASK;
405 			ma += child->bio_ma_n;
406 			if (ma_offset != 0)
407 				ma -= 1;
408 		} else {
409 			child->bio_data = data;
410 			data += child->bio_bcount;
411 		}
412 		cur_offset += child->bio_bcount;
413 		rem_bcount -= child->bio_bcount;
414 	}
415 
416 	return (child_bios);
417 }
418 
419 static int
420 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp,
421     uint32_t alignment)
422 {
423 	struct bio	*child;
424 	struct bio	**child_bios;
425 	int		err, i, num_bios;
426 
427 	child_bios = nvme_construct_child_bios(bp, alignment, &num_bios);
428 	if (child_bios == NULL)
429 		return (ENOMEM);
430 
431 	for (i = 0; i < num_bios; i++) {
432 		child = child_bios[i];
433 		err = nvme_ns_bio_process(ns, child, nvme_bio_child_done);
434 		if (err != 0) {
435 			nvme_bio_child_inbed(bp, err);
436 			g_destroy_bio(child);
437 		}
438 	}
439 
440 	free(child_bios, M_NVME);
441 	return (0);
442 }
443 
444 int
445 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp,
446 	nvme_cb_fn_t cb_fn)
447 {
448 	struct nvme_dsm_range	*dsm_range;
449 	uint32_t		num_bios;
450 	int			err;
451 
452 	bp->bio_driver1 = cb_fn;
453 
454 	if (ns->boundary > 0 &&
455 	    (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) {
456 		num_bios = nvme_get_num_segments(bp->bio_offset,
457 		    bp->bio_bcount, ns->boundary);
458 		if (num_bios > 1)
459 			return (nvme_ns_split_bio(ns, bp, ns->boundary));
460 	}
461 
462 	switch (bp->bio_cmd) {
463 	case BIO_READ:
464 		err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp);
465 		break;
466 	case BIO_WRITE:
467 		err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp);
468 		break;
469 	case BIO_FLUSH:
470 		err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp);
471 		break;
472 	case BIO_DELETE:
473 		dsm_range =
474 		    malloc(sizeof(struct nvme_dsm_range), M_NVME,
475 		    M_ZERO | M_WAITOK);
476 		if (!dsm_range) {
477 			err = ENOMEM;
478 			break;
479 		}
480 		dsm_range->length =
481 		    htole32(bp->bio_bcount/nvme_ns_get_sector_size(ns));
482 		dsm_range->starting_lba =
483 		    htole64(bp->bio_offset/nvme_ns_get_sector_size(ns));
484 		bp->bio_driver2 = dsm_range;
485 		err = nvme_ns_cmd_deallocate(ns, dsm_range, 1,
486 			nvme_ns_bio_done, bp);
487 		if (err != 0)
488 			free(dsm_range, M_NVME);
489 		break;
490 	default:
491 		err = EOPNOTSUPP;
492 		break;
493 	}
494 
495 	return (err);
496 }
497 
498 int
499 nvme_ns_ioctl_process(struct nvme_namespace *ns, u_long cmd, caddr_t arg,
500     int flag, struct thread *td)
501 {
502 	return (nvme_ns_ioctl(ns->cdev, cmd, arg, flag, td));
503 }
504 
505 int
506 nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
507     struct nvme_controller *ctrlr)
508 {
509 	struct make_dev_args                    md_args;
510 	struct nvme_completion_poll_status	status;
511 	int                                     res;
512 	int					unit;
513 	uint8_t					flbas_fmt;
514 	uint8_t					vwc_present;
515 
516 	ns->ctrlr = ctrlr;
517 	ns->id = id;
518 
519 	/*
520 	 * Namespaces are reconstructed after a controller reset, so check
521 	 *  to make sure we only call mtx_init once on each mtx.
522 	 *
523 	 * TODO: Move this somewhere where it gets called at controller
524 	 *  construction time, which is not invoked as part of each
525 	 *  controller reset.
526 	 */
527 	if (!mtx_initialized(&ns->lock))
528 		mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF);
529 
530 	status.done = 0;
531 	nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data,
532 	    nvme_completion_poll_cb, &status);
533 	nvme_completion_poll(&status);
534 	if (nvme_completion_is_error(&status.cpl)) {
535 		nvme_printf(ctrlr, "nvme_identify_namespace failed\n");
536 		return (ENXIO);
537 	}
538 
539 	/* Convert data to host endian */
540 	nvme_namespace_data_swapbytes(&ns->data);
541 
542 	/*
543 	 * If the size of is zero, chances are this isn't a valid
544 	 * namespace (eg one that's not been configured yet). The
545 	 * standard says the entire id will be zeros, so this is a
546 	 * cheap way to test for that.
547 	 */
548 	if (ns->data.nsze == 0)
549 		return (ENXIO);
550 
551 	flbas_fmt = (ns->data.flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) &
552 		NVME_NS_DATA_FLBAS_FORMAT_MASK;
553 	/*
554 	 * Note: format is a 0-based value, so > is appropriate here,
555 	 *  not >=.
556 	 */
557 	if (flbas_fmt > ns->data.nlbaf) {
558 		printf("lba format %d exceeds number supported (%d)\n",
559 		    flbas_fmt, ns->data.nlbaf + 1);
560 		return (ENXIO);
561 	}
562 
563 	/*
564 	 * Older Intel devices advertise in vendor specific space an alignment
565 	 * that improves performance.  If present use for the stripe size.  NVMe
566 	 * 1.3 standardized this as NOIOB, and newer Intel drives use that.
567 	 */
568 	switch (pci_get_devid(ctrlr->dev)) {
569 	case 0x09538086:		/* Intel DC PC3500 */
570 	case 0x0a538086:		/* Intel DC PC3520 */
571 	case 0x0a548086:		/* Intel DC PC4500 */
572 	case 0x0a558086:		/* Dell Intel P4600 */
573 		if (ctrlr->cdata.vs[3] != 0)
574 			ns->boundary =
575 			    (1 << ctrlr->cdata.vs[3]) * ctrlr->min_page_size;
576 		else
577 			ns->boundary = 0;
578 		break;
579 	default:
580 		ns->boundary = ns->data.noiob * nvme_ns_get_sector_size(ns);
581 		break;
582 	}
583 
584 	if (nvme_ctrlr_has_dataset_mgmt(&ctrlr->cdata))
585 		ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED;
586 
587 	vwc_present = (ctrlr->cdata.vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) &
588 		NVME_CTRLR_DATA_VWC_PRESENT_MASK;
589 	if (vwc_present)
590 		ns->flags |= NVME_NS_FLUSH_SUPPORTED;
591 
592 	/*
593 	 * cdev may have already been created, if we are reconstructing the
594 	 *  namespace after a controller-level reset.
595 	 */
596 	if (ns->cdev != NULL)
597 		return (0);
598 
599 	/*
600 	 * Namespace IDs start at 1, so we need to subtract 1 to create a
601 	 *  correct unit number.
602 	 */
603 	unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1;
604 
605 	make_dev_args_init(&md_args);
606 	md_args.mda_devsw = &nvme_ns_cdevsw;
607 	md_args.mda_unit = unit;
608 	md_args.mda_mode = 0600;
609 	md_args.mda_si_drv1 = ns;
610 	res = make_dev_s(&md_args, &ns->cdev, "nvme%dns%d",
611 	    device_get_unit(ctrlr->dev), ns->id);
612 	if (res != 0)
613 		return (ENXIO);
614 
615 	ns->cdev->si_flags |= SI_UNMAPPED;
616 
617 	return (0);
618 }
619 
620 void nvme_ns_destruct(struct nvme_namespace *ns)
621 {
622 
623 	if (ns->cdev != NULL)
624 		destroy_dev(ns->cdev);
625 }
626