xref: /freebsd/sys/dev/nvme/nvme_ns.c (revision 4232f826683298e85d469c0ef17259cecd13b2c7) 
1 /*-
2  * Copyright (C) 2012-2013 Intel Corporation
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/bus.h>
33 #include <sys/conf.h>
34 #include <sys/disk.h>
35 #include <sys/fcntl.h>
36 #include <sys/ioccom.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/proc.h>
40 #include <sys/systm.h>
41 
42 #include <dev/pci/pcivar.h>
43 
44 #include <geom/geom.h>
45 
46 #include "nvme_private.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 DIOCGMEDIASIZE:
84 		*(off_t *)arg = (off_t)nvme_ns_get_size(ns);
85 		break;
86 	case DIOCGSECTORSIZE:
87 		*(u_int *)arg = nvme_ns_get_sector_size(ns);
88 		break;
89 	default:
90 		return (ENOTTY);
91 	}
92 
93 	return (0);
94 }
95 
96 static int
97 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused,
98     struct thread *td)
99 {
100 	int error = 0;
101 
102 	if (flags & FWRITE)
103 		error = securelevel_gt(td->td_ucred, 0);
104 
105 	return (error);
106 }
107 
108 static int
109 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused,
110     struct thread *td)
111 {
112 
113 	return (0);
114 }
115 
116 static void
117 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl)
118 {
119 	struct bio *bp = arg;
120 
121 	/*
122 	 * TODO: add more extensive translation of NVMe status codes
123 	 *  to different bio error codes (i.e. EIO, EINVAL, etc.)
124 	 */
125 	if (nvme_completion_is_error(cpl)) {
126 		bp->bio_error = EIO;
127 		bp->bio_flags |= BIO_ERROR;
128 		bp->bio_resid = bp->bio_bcount;
129 	} else
130 		bp->bio_resid = 0;
131 
132 	biodone(bp);
133 }
134 
135 static void
136 nvme_ns_strategy(struct bio *bp)
137 {
138 	struct nvme_namespace	*ns;
139 	int			err;
140 
141 	ns = bp->bio_dev->si_drv1;
142 	err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done);
143 
144 	if (err) {
145 		bp->bio_error = err;
146 		bp->bio_flags |= BIO_ERROR;
147 		bp->bio_resid = bp->bio_bcount;
148 		biodone(bp);
149 	}
150 
151 }
152 
153 static struct cdevsw nvme_ns_cdevsw = {
154 	.d_version =	D_VERSION,
155 	.d_flags =	D_DISK,
156 	.d_read =	physread,
157 	.d_write =	physwrite,
158 	.d_open =	nvme_ns_open,
159 	.d_close =	nvme_ns_close,
160 	.d_strategy =	nvme_ns_strategy,
161 	.d_ioctl =	nvme_ns_ioctl
162 };
163 
164 uint32_t
165 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns)
166 {
167 	return ns->ctrlr->max_xfer_size;
168 }
169 
170 uint32_t
171 nvme_ns_get_sector_size(struct nvme_namespace *ns)
172 {
173 	return (1 << ns->data.lbaf[ns->data.flbas.format].lbads);
174 }
175 
176 uint64_t
177 nvme_ns_get_num_sectors(struct nvme_namespace *ns)
178 {
179 	return (ns->data.nsze);
180 }
181 
182 uint64_t
183 nvme_ns_get_size(struct nvme_namespace *ns)
184 {
185 	return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns));
186 }
187 
188 uint32_t
189 nvme_ns_get_flags(struct nvme_namespace *ns)
190 {
191 	return (ns->flags);
192 }
193 
194 const char *
195 nvme_ns_get_serial_number(struct nvme_namespace *ns)
196 {
197 	return ((const char *)ns->ctrlr->cdata.sn);
198 }
199 
200 const char *
201 nvme_ns_get_model_number(struct nvme_namespace *ns)
202 {
203 	return ((const char *)ns->ctrlr->cdata.mn);
204 }
205 
206 const struct nvme_namespace_data *
207 nvme_ns_get_data(struct nvme_namespace *ns)
208 {
209 
210 	return (&ns->data);
211 }
212 
213 static void
214 nvme_ns_bio_done(void *arg, const struct nvme_completion *status)
215 {
216 	struct bio	*bp = arg;
217 	nvme_cb_fn_t	bp_cb_fn;
218 
219 	bp_cb_fn = bp->bio_driver1;
220 
221 	if (bp->bio_driver2)
222 		free(bp->bio_driver2, M_NVME);
223 
224 	if (nvme_completion_is_error(status)) {
225 		bp->bio_flags |= BIO_ERROR;
226 		if (bp->bio_error == 0)
227 			bp->bio_error = EIO;
228 	}
229 
230 	if ((bp->bio_flags & BIO_ERROR) == 0)
231 		bp->bio_resid = 0;
232 	else
233 		bp->bio_resid = bp->bio_bcount;
234 
235 	bp_cb_fn(bp, status);
236 }
237 
238 static void
239 nvme_bio_child_inbed(struct bio *parent, int bio_error)
240 {
241 	struct nvme_completion	parent_cpl;
242 	int			inbed;
243 
244 	if (bio_error != 0) {
245 		parent->bio_flags |= BIO_ERROR;
246 		parent->bio_error = bio_error;
247 	}
248 
249 	/*
250 	 * atomic_fetchadd will return value before adding 1, so we still
251 	 *  must add 1 to get the updated inbed number.
252 	 */
253 	inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1;
254 	if (inbed == parent->bio_children) {
255 		bzero(&parent_cpl, sizeof(parent_cpl));
256 		if (parent->bio_flags & BIO_ERROR)
257 			parent_cpl.status.sc = NVME_SC_DATA_TRANSFER_ERROR;
258 		nvme_ns_bio_done(parent, &parent_cpl);
259 	}
260 }
261 
262 static void
263 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl)
264 {
265 	struct bio		*child = arg;
266 	struct bio		*parent;
267 	int			bio_error;
268 
269 	parent = child->bio_parent;
270 	g_destroy_bio(child);
271 	bio_error = nvme_completion_is_error(cpl) ? EIO : 0;
272 	nvme_bio_child_inbed(parent, bio_error);
273 }
274 
275 static uint32_t
276 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align)
277 {
278 	uint32_t	num_segs, offset, remainder;
279 
280 	if (align == 0)
281 		return (1);
282 
283 	KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n"));
284 
285 	num_segs = size / align;
286 	remainder = size & (align - 1);
287 	offset = addr & (align - 1);
288 	if (remainder > 0 || offset > 0)
289 		num_segs += 1 + (remainder + offset - 1) / align;
290 	return (num_segs);
291 }
292 
293 static void
294 nvme_free_child_bios(int num_bios, struct bio **child_bios)
295 {
296 	int i;
297 
298 	for (i = 0; i < num_bios; i++) {
299 		if (child_bios[i] != NULL)
300 			g_destroy_bio(child_bios[i]);
301 	}
302 
303 	free(child_bios, M_NVME);
304 }
305 
306 static struct bio **
307 nvme_allocate_child_bios(int num_bios)
308 {
309 	struct bio **child_bios;
310 	int err = 0, i;
311 
312 	child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT);
313 	if (child_bios == NULL)
314 		return (NULL);
315 
316 	for (i = 0; i < num_bios; i++) {
317 		child_bios[i] = g_new_bio();
318 		if (child_bios[i] == NULL)
319 			err = ENOMEM;
320 	}
321 
322 	if (err == ENOMEM) {
323 		nvme_free_child_bios(num_bios, child_bios);
324 		return (NULL);
325 	}
326 
327 	return (child_bios);
328 }
329 
330 static struct bio **
331 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios)
332 {
333 	struct bio	**child_bios;
334 	struct bio	*child;
335 	uint64_t	cur_offset;
336 	caddr_t		data;
337 	uint32_t	rem_bcount;
338 	int		i;
339 #ifdef NVME_UNMAPPED_BIO_SUPPORT
340 	struct vm_page	**ma;
341 	uint32_t	ma_offset;
342 #endif
343 
344 	*num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount,
345 	    alignment);
346 	child_bios = nvme_allocate_child_bios(*num_bios);
347 	if (child_bios == NULL)
348 		return (NULL);
349 
350 	bp->bio_children = *num_bios;
351 	bp->bio_inbed = 0;
352 	cur_offset = bp->bio_offset;
353 	rem_bcount = bp->bio_bcount;
354 	data = bp->bio_data;
355 #ifdef NVME_UNMAPPED_BIO_SUPPORT
356 	ma_offset = bp->bio_ma_offset;
357 	ma = bp->bio_ma;
358 #endif
359 
360 	for (i = 0; i < *num_bios; i++) {
361 		child = child_bios[i];
362 		child->bio_parent = bp;
363 		child->bio_cmd = bp->bio_cmd;
364 		child->bio_offset = cur_offset;
365 		child->bio_bcount = min(rem_bcount,
366 		    alignment - (cur_offset & (alignment - 1)));
367 		child->bio_flags = bp->bio_flags;
368 #ifdef NVME_UNMAPPED_BIO_SUPPORT
369 		if (bp->bio_flags & BIO_UNMAPPED) {
370 			child->bio_ma_offset = ma_offset;
371 			child->bio_ma = ma;
372 			child->bio_ma_n =
373 			    nvme_get_num_segments(child->bio_ma_offset,
374 				child->bio_bcount, PAGE_SIZE);
375 			ma_offset = (ma_offset + child->bio_bcount) &
376 			    PAGE_MASK;
377 			ma += child->bio_ma_n;
378 			if (ma_offset != 0)
379 				ma -= 1;
380 		} else
381 #endif
382 		{
383 			child->bio_data = data;
384 			data += child->bio_bcount;
385 		}
386 		cur_offset += child->bio_bcount;
387 		rem_bcount -= child->bio_bcount;
388 	}
389 
390 	return (child_bios);
391 }
392 
393 static int
394 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp,
395     uint32_t alignment)
396 {
397 	struct bio	*child;
398 	struct bio	**child_bios;
399 	int		err, i, num_bios;
400 
401 	child_bios = nvme_construct_child_bios(bp, alignment, &num_bios);
402 	if (child_bios == NULL)
403 		return (ENOMEM);
404 
405 	for (i = 0; i < num_bios; i++) {
406 		child = child_bios[i];
407 		err = nvme_ns_bio_process(ns, child, nvme_bio_child_done);
408 		if (err != 0) {
409 			nvme_bio_child_inbed(bp, err);
410 			g_destroy_bio(child);
411 		}
412 	}
413 
414 	free(child_bios, M_NVME);
415 	return (0);
416 }
417 
418 int
419 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp,
420 	nvme_cb_fn_t cb_fn)
421 {
422 	struct nvme_dsm_range	*dsm_range;
423 	uint32_t		num_bios;
424 	int			err;
425 
426 	bp->bio_driver1 = cb_fn;
427 
428 	if (ns->stripesize > 0 &&
429 	    (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) {
430 		num_bios = nvme_get_num_segments(bp->bio_offset,
431 		    bp->bio_bcount, ns->stripesize);
432 		if (num_bios > 1)
433 			return (nvme_ns_split_bio(ns, bp, ns->stripesize));
434 	}
435 
436 	switch (bp->bio_cmd) {
437 	case BIO_READ:
438 		err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp);
439 		break;
440 	case BIO_WRITE:
441 		err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp);
442 		break;
443 	case BIO_FLUSH:
444 		err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp);
445 		break;
446 	case BIO_DELETE:
447 		dsm_range =
448 		    malloc(sizeof(struct nvme_dsm_range), M_NVME,
449 		    M_ZERO | M_WAITOK);
450 		dsm_range->length =
451 		    bp->bio_bcount/nvme_ns_get_sector_size(ns);
452 		dsm_range->starting_lba =
453 		    bp->bio_offset/nvme_ns_get_sector_size(ns);
454 		bp->bio_driver2 = dsm_range;
455 		err = nvme_ns_cmd_deallocate(ns, dsm_range, 1,
456 			nvme_ns_bio_done, bp);
457 		if (err != 0)
458 			free(dsm_range, M_NVME);
459 		break;
460 	default:
461 		err = EIO;
462 		break;
463 	}
464 
465 	return (err);
466 }
467 
468 int
469 nvme_ns_construct(struct nvme_namespace *ns, uint16_t id,
470     struct nvme_controller *ctrlr)
471 {
472 	struct nvme_completion_poll_status	status;
473 	int					unit;
474 
475 	ns->ctrlr = ctrlr;
476 	ns->id = id;
477 	ns->stripesize = 0;
478 
479 	if (pci_get_devid(ctrlr->dev) == 0x09538086 && ctrlr->cdata.vs[3] != 0)
480 		ns->stripesize =
481 		    (1 << ctrlr->cdata.vs[3]) * ctrlr->min_page_size;
482 
483 	/*
484 	 * Namespaces are reconstructed after a controller reset, so check
485 	 *  to make sure we only call mtx_init once on each mtx.
486 	 *
487 	 * TODO: Move this somewhere where it gets called at controller
488 	 *  construction time, which is not invoked as part of each
489 	 *  controller reset.
490 	 */
491 	if (!mtx_initialized(&ns->lock))
492 		mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF);
493 
494 	status.done = FALSE;
495 	nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data,
496 	    nvme_completion_poll_cb, &status);
497 	while (status.done == FALSE)
498 		DELAY(5);
499 	if (nvme_completion_is_error(&status.cpl)) {
500 		nvme_printf(ctrlr, "nvme_identify_namespace failed\n");
501 		return (ENXIO);
502 	}
503 
504 	/*
505 	 * Note: format is a 0-based value, so > is appropriate here,
506 	 *  not >=.
507 	 */
508 	if (ns->data.flbas.format > ns->data.nlbaf) {
509 		printf("lba format %d exceeds number supported (%d)\n",
510 		    ns->data.flbas.format, ns->data.nlbaf+1);
511 		return (1);
512 	}
513 
514 	if (ctrlr->cdata.oncs.dsm)
515 		ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED;
516 
517 	if (ctrlr->cdata.vwc.present)
518 		ns->flags |= NVME_NS_FLUSH_SUPPORTED;
519 
520 	/*
521 	 * cdev may have already been created, if we are reconstructing the
522 	 *  namespace after a controller-level reset.
523 	 */
524 	if (ns->cdev != NULL)
525 		return (0);
526 
527 	/*
528 	 * Namespace IDs start at 1, so we need to subtract 1 to create a
529 	 *  correct unit number.
530 	 */
531 	unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1;
532 
533 /*
534  * MAKEDEV_ETERNAL was added in r210923, for cdevs that will never
535  *  be destroyed.  This avoids refcounting on the cdev object.
536  *  That should be OK case here, as long as we're not supporting PCIe
537  *  surprise removal nor namespace deletion.
538  */
539 #ifdef MAKEDEV_ETERNAL_KLD
540 	ns->cdev = make_dev_credf(MAKEDEV_ETERNAL_KLD, &nvme_ns_cdevsw, unit,
541 	    NULL, UID_ROOT, GID_WHEEL, 0600, "nvme%dns%d",
542 	    device_get_unit(ctrlr->dev), ns->id);
543 #else
544 	ns->cdev = make_dev_credf(0, &nvme_ns_cdevsw, unit,
545 	    NULL, UID_ROOT, GID_WHEEL, 0600, "nvme%dns%d",
546 	    device_get_unit(ctrlr->dev), ns->id);
547 #endif
548 #ifdef NVME_UNMAPPED_BIO_SUPPORT
549 	ns->cdev->si_flags |= SI_UNMAPPED;
550 #endif
551 
552 	if (ns->cdev != NULL)
553 		ns->cdev->si_drv1 = ns;
554 
555 	return (0);
556 }
557 
558 void nvme_ns_destruct(struct nvme_namespace *ns)
559 {
560 
561 	if (ns->cdev != NULL)
562 		destroy_dev(ns->cdev);
563 }
564