xref: /freebsd/sys/dev/nvme/nvme_ns.c (revision 1ae25866767d686067fe6678b62681b7a8f0d361) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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/param.h>
30 #include <sys/bio.h>
31 #include <sys/bus.h>
32 #include <sys/conf.h>
33 #include <sys/disk.h>
34 #include <sys/fcntl.h>
35 #include <sys/ioccom.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/proc.h>
39 #include <sys/systm.h>
40 
41 #include <dev/pci/pcivar.h>
42 
43 #include <geom/geom.h>
44 
45 #include "nvme_private.h"
46 #include "nvme_linux.h"
47 
48 static void		nvme_bio_child_inbed(struct bio *parent, int bio_error);
49 static void		nvme_bio_child_done(void *arg,
50 					    const struct nvme_completion *cpl);
51 static uint32_t		nvme_get_num_segments(uint64_t addr, uint64_t size,
52 					      uint32_t alignment);
53 static void		nvme_free_child_bios(int num_bios,
54 					     struct bio **child_bios);
55 static struct bio **	nvme_allocate_child_bios(int num_bios);
56 static struct bio **	nvme_construct_child_bios(struct bio *bp,
57 						  uint32_t alignment,
58 						  int *num_bios);
59 static int		nvme_ns_split_bio(struct nvme_namespace *ns,
60 					  struct bio *bp,
61 					  uint32_t alignment);
62 
63 static int
64 nvme_ns_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag,
65     struct thread *td)
66 {
67 	struct nvme_namespace			*ns;
68 	struct nvme_controller			*ctrlr;
69 	struct nvme_pt_command			*pt;
70 
71 	ns = cdev->si_drv1;
72 	ctrlr = ns->ctrlr;
73 
74 	switch (cmd) {
75 	case NVME_IO_TEST:
76 	case NVME_BIO_TEST:
77 		nvme_ns_test(ns, cmd, arg);
78 		break;
79 	case NVME_PASSTHROUGH_CMD:
80 		pt = (struct nvme_pt_command *)arg;
81 		return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, ns->id,
82 		    1 /* is_user_buffer */, 0 /* is_admin_cmd */));
83 	case NVME_GET_NSID:
84 	{
85 		struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)arg;
86 		strlcpy(gnsid->cdev, device_get_nameunit(ctrlr->dev),
87 		    sizeof(gnsid->cdev));
88 		gnsid->nsid = ns->id;
89 		break;
90 	}
91 	case DIOCGIDENT: {
92 		uint8_t *sn = arg;
93 		nvme_ctrlr_get_ident(ctrlr, sn);
94 		break;
95 	}
96 	case DIOCGMEDIASIZE:
97 		*(off_t *)arg = (off_t)nvme_ns_get_size(ns);
98 		break;
99 	case DIOCGSECTORSIZE:
100 		*(u_int *)arg = nvme_ns_get_sector_size(ns);
101 		break;
102 	/* Linux Compatible (see nvme_linux.h) */
103 	case NVME_IOCTL_ID:
104 		td->td_retval[0] = ns->id;
105 		return (0);
106 
107 	case NVME_IOCTL_ADMIN_CMD:
108 	case NVME_IOCTL_IO_CMD: {
109 		struct nvme_passthru_cmd *npc = (struct nvme_passthru_cmd *)arg;
110 
111 		return (nvme_ctrlr_linux_passthru_cmd(ctrlr, npc, ns->id, true,
112 		    cmd == NVME_IOCTL_ADMIN_CMD));
113 	}
114 	default:
115 		return (ENOTTY);
116 	}
117 
118 	return (0);
119 }
120 
121 static int
122 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused,
123     struct thread *td)
124 {
125 	int error = 0;
126 
127 	if (flags & FWRITE)
128 		error = securelevel_gt(td->td_ucred, 0);
129 
130 	return (error);
131 }
132 
133 static int
134 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused,
135     struct thread *td)
136 {
137 	return (0);
138 }
139 
140 static void
141 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl)
142 {
143 	struct bio *bp = arg;
144 
145 	/*
146 	 * TODO: add more extensive translation of NVMe status codes
147 	 *  to different bio error codes (i.e. EIO, EINVAL, etc.)
148 	 */
149 	if (nvme_completion_is_error(cpl)) {
150 		bp->bio_error = EIO;
151 		bp->bio_flags |= BIO_ERROR;
152 		bp->bio_resid = bp->bio_bcount;
153 	} else
154 		bp->bio_resid = 0;
155 
156 	biodone(bp);
157 }
158 
159 static void
160 nvme_ns_strategy(struct bio *bp)
161 {
162 	struct nvme_namespace	*ns;
163 	int			err;
164 
165 	ns = bp->bio_dev->si_drv1;
166 	err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done);
167 
168 	if (err) {
169 		bp->bio_error = err;
170 		bp->bio_flags |= BIO_ERROR;
171 		bp->bio_resid = bp->bio_bcount;
172 		biodone(bp);
173 	}
174 
175 }
176 
177 static struct cdevsw nvme_ns_cdevsw = {
178 	.d_version =	D_VERSION,
179 	.d_flags =	D_DISK,
180 	.d_read =	physread,
181 	.d_write =	physwrite,
182 	.d_open =	nvme_ns_open,
183 	.d_close =	nvme_ns_close,
184 	.d_strategy =	nvme_ns_strategy,
185 	.d_ioctl =	nvme_ns_ioctl
186 };
187 
188 uint32_t
189 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns)
190 {
191 	return ns->ctrlr->max_xfer_size;
192 }
193 
194 uint32_t
195 nvme_ns_get_sector_size(struct nvme_namespace *ns)
196 {
197 	uint8_t flbas_fmt, lbads;
198 
199 	flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas);
200 	lbads = NVMEV(NVME_NS_DATA_LBAF_LBADS, ns->data.lbaf[flbas_fmt]);
201 
202 	return (1 << lbads);
203 }
204 
205 uint64_t
206 nvme_ns_get_num_sectors(struct nvme_namespace *ns)
207 {
208 	return (ns->data.nsze);
209 }
210 
211 uint64_t
212 nvme_ns_get_size(struct nvme_namespace *ns)
213 {
214 	return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns));
215 }
216 
217 uint32_t
218 nvme_ns_get_flags(struct nvme_namespace *ns)
219 {
220 	return (ns->flags);
221 }
222 
223 const char *
224 nvme_ns_get_serial_number(struct nvme_namespace *ns)
225 {
226 	return ((const char *)ns->ctrlr->cdata.sn);
227 }
228 
229 const char *
230 nvme_ns_get_model_number(struct nvme_namespace *ns)
231 {
232 	return ((const char *)ns->ctrlr->cdata.mn);
233 }
234 
235 const struct nvme_namespace_data *
236 nvme_ns_get_data(struct nvme_namespace *ns)
237 {
238 	return (&ns->data);
239 }
240 
241 uint32_t
242 nvme_ns_get_stripesize(struct nvme_namespace *ns)
243 {
244 	uint32_t ss;
245 
246 	if (NVMEV(NVME_NS_DATA_NSFEAT_NPVALID, ns->data.nsfeat) != 0) {
247 		ss = nvme_ns_get_sector_size(ns);
248 		if (ns->data.npwa != 0)
249 			return ((ns->data.npwa + 1) * ss);
250 		else if (ns->data.npwg != 0)
251 			return ((ns->data.npwg + 1) * ss);
252 	}
253 	return (ns->boundary);
254 }
255 
256 static void
257 nvme_ns_bio_done(void *arg, const struct nvme_completion *status)
258 {
259 	struct bio	*bp = arg;
260 	nvme_cb_fn_t	bp_cb_fn;
261 
262 	bp_cb_fn = bp->bio_driver1;
263 
264 	if (bp->bio_driver2)
265 		free(bp->bio_driver2, M_NVME);
266 
267 	if (nvme_completion_is_error(status)) {
268 		bp->bio_flags |= BIO_ERROR;
269 		if (bp->bio_error == 0)
270 			bp->bio_error = EIO;
271 	}
272 
273 	if ((bp->bio_flags & BIO_ERROR) == 0)
274 		bp->bio_resid = 0;
275 	else
276 		bp->bio_resid = bp->bio_bcount;
277 
278 	bp_cb_fn(bp, status);
279 }
280 
281 static void
282 nvme_bio_child_inbed(struct bio *parent, int bio_error)
283 {
284 	struct nvme_completion	parent_cpl;
285 	int			children, inbed;
286 
287 	if (bio_error != 0) {
288 		parent->bio_flags |= BIO_ERROR;
289 		parent->bio_error = bio_error;
290 	}
291 
292 	/*
293 	 * atomic_fetchadd will return value before adding 1, so we still
294 	 *  must add 1 to get the updated inbed number.  Save bio_children
295 	 *  before incrementing to guard against race conditions when
296 	 *  two children bios complete on different queues.
297 	 */
298 	children = atomic_load_acq_int(&parent->bio_children);
299 	inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1;
300 	if (inbed == children) {
301 		bzero(&parent_cpl, sizeof(parent_cpl));
302 		if (parent->bio_flags & BIO_ERROR) {
303 			parent_cpl.status &= ~NVMEM(NVME_STATUS_SC);
304 			parent_cpl.status |= NVMEF(NVME_STATUS_SC,
305 			    NVME_SC_DATA_TRANSFER_ERROR);
306 		}
307 		nvme_ns_bio_done(parent, &parent_cpl);
308 	}
309 }
310 
311 static void
312 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl)
313 {
314 	struct bio		*child = arg;
315 	struct bio		*parent;
316 	int			bio_error;
317 
318 	parent = child->bio_parent;
319 	g_destroy_bio(child);
320 	bio_error = nvme_completion_is_error(cpl) ? EIO : 0;
321 	nvme_bio_child_inbed(parent, bio_error);
322 }
323 
324 static uint32_t
325 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align)
326 {
327 	uint32_t	num_segs, offset, remainder;
328 
329 	if (align == 0)
330 		return (1);
331 
332 	KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n"));
333 
334 	num_segs = size / align;
335 	remainder = size & (align - 1);
336 	offset = addr & (align - 1);
337 	if (remainder > 0 || offset > 0)
338 		num_segs += 1 + (remainder + offset - 1) / align;
339 	return (num_segs);
340 }
341 
342 static void
343 nvme_free_child_bios(int num_bios, struct bio **child_bios)
344 {
345 	int i;
346 
347 	for (i = 0; i < num_bios; i++) {
348 		if (child_bios[i] != NULL)
349 			g_destroy_bio(child_bios[i]);
350 	}
351 
352 	free(child_bios, M_NVME);
353 }
354 
355 static struct bio **
356 nvme_allocate_child_bios(int num_bios)
357 {
358 	struct bio **child_bios;
359 	int err = 0, i;
360 
361 	child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT);
362 	if (child_bios == NULL)
363 		return (NULL);
364 
365 	for (i = 0; i < num_bios; i++) {
366 		child_bios[i] = g_new_bio();
367 		if (child_bios[i] == NULL)
368 			err = ENOMEM;
369 	}
370 
371 	if (err == ENOMEM) {
372 		nvme_free_child_bios(num_bios, child_bios);
373 		return (NULL);
374 	}
375 
376 	return (child_bios);
377 }
378 
379 static struct bio **
380 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios)
381 {
382 	struct bio	**child_bios;
383 	struct bio	*child;
384 	uint64_t	cur_offset;
385 	caddr_t		data;
386 	uint32_t	rem_bcount;
387 	int		i;
388 	struct vm_page	**ma;
389 	uint32_t	ma_offset;
390 
391 	*num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount,
392 	    alignment);
393 	child_bios = nvme_allocate_child_bios(*num_bios);
394 	if (child_bios == NULL)
395 		return (NULL);
396 
397 	bp->bio_children = *num_bios;
398 	bp->bio_inbed = 0;
399 	cur_offset = bp->bio_offset;
400 	rem_bcount = bp->bio_bcount;
401 	data = bp->bio_data;
402 	ma_offset = bp->bio_ma_offset;
403 	ma = bp->bio_ma;
404 
405 	for (i = 0; i < *num_bios; i++) {
406 		child = child_bios[i];
407 		child->bio_parent = bp;
408 		child->bio_cmd = bp->bio_cmd;
409 		child->bio_offset = cur_offset;
410 		child->bio_bcount = min(rem_bcount,
411 		    alignment - (cur_offset & (alignment - 1)));
412 		child->bio_flags = bp->bio_flags;
413 		if (bp->bio_flags & BIO_UNMAPPED) {
414 			child->bio_ma_offset = ma_offset;
415 			child->bio_ma = ma;
416 			child->bio_ma_n =
417 			    nvme_get_num_segments(child->bio_ma_offset,
418 				child->bio_bcount, PAGE_SIZE);
419 			ma_offset = (ma_offset + child->bio_bcount) &
420 			    PAGE_MASK;
421 			ma += child->bio_ma_n;
422 			if (ma_offset != 0)
423 				ma -= 1;
424 		} else {
425 			child->bio_data = data;
426 			data += child->bio_bcount;
427 		}
428 		cur_offset += child->bio_bcount;
429 		rem_bcount -= child->bio_bcount;
430 	}
431 
432 	return (child_bios);
433 }
434 
435 static int
436 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp,
437     uint32_t alignment)
438 {
439 	struct bio	*child;
440 	struct bio	**child_bios;
441 	int		err, i, num_bios;
442 
443 	child_bios = nvme_construct_child_bios(bp, alignment, &num_bios);
444 	if (child_bios == NULL)
445 		return (ENOMEM);
446 
447 	counter_u64_add(ns->ctrlr->alignment_splits, 1);
448 	for (i = 0; i < num_bios; i++) {
449 		child = child_bios[i];
450 		err = nvme_ns_bio_process(ns, child, nvme_bio_child_done);
451 		if (err != 0) {
452 			nvme_bio_child_inbed(bp, err);
453 			g_destroy_bio(child);
454 		}
455 	}
456 
457 	free(child_bios, M_NVME);
458 	return (0);
459 }
460 
461 int
462 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp,
463 	nvme_cb_fn_t cb_fn)
464 {
465 	struct nvme_dsm_range	*dsm_range;
466 	uint32_t		num_bios;
467 	int			err;
468 
469 	bp->bio_driver1 = cb_fn;
470 
471 	if (ns->boundary > 0 &&
472 	    (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) {
473 		num_bios = nvme_get_num_segments(bp->bio_offset,
474 		    bp->bio_bcount, ns->boundary);
475 		if (num_bios > 1)
476 			return (nvme_ns_split_bio(ns, bp, ns->boundary));
477 	}
478 
479 	switch (bp->bio_cmd) {
480 	case BIO_READ:
481 		err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp);
482 		break;
483 	case BIO_WRITE:
484 		err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp);
485 		break;
486 	case BIO_FLUSH:
487 		err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp);
488 		break;
489 	case BIO_DELETE:
490 		dsm_range =
491 		    malloc(sizeof(struct nvme_dsm_range), M_NVME,
492 		    M_ZERO | M_NOWAIT);
493 		if (!dsm_range) {
494 			err = ENOMEM;
495 			break;
496 		}
497 		dsm_range->length =
498 		    htole32(bp->bio_bcount/nvme_ns_get_sector_size(ns));
499 		dsm_range->starting_lba =
500 		    htole64(bp->bio_offset/nvme_ns_get_sector_size(ns));
501 		bp->bio_driver2 = dsm_range;
502 		err = nvme_ns_cmd_deallocate(ns, dsm_range, 1,
503 			nvme_ns_bio_done, bp);
504 		if (err != 0)
505 			free(dsm_range, M_NVME);
506 		break;
507 	default:
508 		err = EOPNOTSUPP;
509 		break;
510 	}
511 
512 	return (err);
513 }
514 
515 int
516 nvme_ns_ioctl_process(struct nvme_namespace *ns, u_long cmd, caddr_t arg,
517     int flag, struct thread *td)
518 {
519 	return (nvme_ns_ioctl(ns->cdev, cmd, arg, flag, td));
520 }
521 
522 int
523 nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
524     struct nvme_controller *ctrlr)
525 {
526 	struct make_dev_args                    md_args;
527 	struct nvme_completion_poll_status	status;
528 	int                                     res;
529 	int					unit;
530 	uint8_t					flbas_fmt;
531 	uint8_t					vwc_present;
532 
533 	ns->ctrlr = ctrlr;
534 	ns->id = id;
535 
536 	/*
537 	 * Namespaces are reconstructed after a controller reset, so check
538 	 *  to make sure we only call mtx_init once on each mtx.
539 	 *
540 	 * TODO: Move this somewhere where it gets called at controller
541 	 *  construction time, which is not invoked as part of each
542 	 *  controller reset.
543 	 */
544 	if (!mtx_initialized(&ns->lock))
545 		mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF);
546 
547 	status.done = 0;
548 	nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data,
549 	    nvme_completion_poll_cb, &status);
550 	nvme_completion_poll(&status);
551 	if (nvme_completion_is_error(&status.cpl)) {
552 		nvme_printf(ctrlr, "nvme_identify_namespace failed\n");
553 		return (ENXIO);
554 	}
555 
556 	/* Convert data to host endian */
557 	nvme_namespace_data_swapbytes(&ns->data);
558 
559 	/*
560 	 * If the size of is zero, chances are this isn't a valid
561 	 * namespace (eg one that's not been configured yet). The
562 	 * standard says the entire id will be zeros, so this is a
563 	 * cheap way to test for that.
564 	 */
565 	if (ns->data.nsze == 0)
566 		return (ENXIO);
567 
568 	flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas);
569 
570 	/*
571 	 * Note: format is a 0-based value, so > is appropriate here,
572 	 *  not >=.
573 	 */
574 	if (flbas_fmt > ns->data.nlbaf) {
575 		nvme_printf(ctrlr,
576 		    "lba format %d exceeds number supported (%d)\n",
577 		    flbas_fmt, ns->data.nlbaf + 1);
578 		return (ENXIO);
579 	}
580 
581 	/*
582 	 * Older Intel devices (like the PC35xxx and P45xx series) advertise in
583 	 * vendor specific space an alignment that improves performance.  If
584 	 * present use for the stripe size.  NVMe 1.3 standardized this as
585 	 * NOIOB, and newer Intel drives use that.
586 	 */
587 	if ((ctrlr->quirks & QUIRK_INTEL_ALIGNMENT) != 0) {
588 		if (ctrlr->cdata.vs[3] != 0)
589 			ns->boundary =
590 			    1 << (ctrlr->cdata.vs[3] + NVME_MPS_SHIFT +
591 				NVME_CAP_HI_MPSMIN(ctrlr->cap_hi));
592 		else
593 			ns->boundary = 0;
594 	} else {
595 		ns->boundary = ns->data.noiob * nvme_ns_get_sector_size(ns);
596 	}
597 
598 	if (nvme_ctrlr_has_dataset_mgmt(&ctrlr->cdata))
599 		ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED;
600 
601 	vwc_present = NVMEV(NVME_CTRLR_DATA_VWC_PRESENT, ctrlr->cdata.vwc);
602 	if (vwc_present)
603 		ns->flags |= NVME_NS_FLUSH_SUPPORTED;
604 
605 	/*
606 	 * cdev may have already been created, if we are reconstructing the
607 	 *  namespace after a controller-level reset.
608 	 */
609 	if (ns->cdev != NULL)
610 		return (0);
611 
612 	/*
613 	 * Namespace IDs start at 1, so we need to subtract 1 to create a
614 	 *  correct unit number.
615 	 */
616 	unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1;
617 
618 	make_dev_args_init(&md_args);
619 	md_args.mda_devsw = &nvme_ns_cdevsw;
620 	md_args.mda_unit = unit;
621 	md_args.mda_mode = 0600;
622 	md_args.mda_si_drv1 = ns;
623 	res = make_dev_s(&md_args, &ns->cdev, "%sn%d",
624 	    device_get_nameunit(ctrlr->dev), ns->id);
625 	if (res != 0)
626 		return (ENXIO);
627 	ns->cdev->si_drv2 = make_dev_alias(ns->cdev, "%sns%d",
628 	    device_get_nameunit(ctrlr->dev), ns->id);
629 	ns->cdev->si_flags |= SI_UNMAPPED;
630 
631 	return (0);
632 }
633 
634 void
635 nvme_ns_destruct(struct nvme_namespace *ns)
636 {
637 	if (ns->cdev != NULL) {
638 		if (ns->cdev->si_drv2 != NULL)
639 			destroy_dev(ns->cdev->si_drv2);
640 		destroy_dev(ns->cdev);
641 	}
642 }
643