1// SPDX-License-Identifier: GPL-2.0
2/*
3 * NVMe I/O command implementation.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5 */
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7#include <linux/blkdev.h>
8#include <linux/blk-integrity.h>
9#include <linux/memremap.h>
10#include <linux/module.h>
11#include "nvmet.h"
12
13void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)
14{
15	/* Logical blocks per physical block, 0's based. */
16	const __le16 lpp0b = to0based(bdev_physical_block_size(bdev) /
17				      bdev_logical_block_size(bdev));
18
19	/*
20	 * For NVMe 1.2 and later, bit 1 indicates that the fields NAWUN,
21	 * NAWUPF, and NACWU are defined for this namespace and should be
22	 * used by the host for this namespace instead of the AWUN, AWUPF,
23	 * and ACWU fields in the Identify Controller data structure. If
24	 * any of these fields are zero that means that the corresponding
25	 * field from the identify controller data structure should be used.
26	 */
27	id->nsfeat |= 1 << 1;
28	id->nawun = lpp0b;
29	id->nawupf = lpp0b;
30	id->nacwu = lpp0b;
31
32	/*
33	 * Bit 4 indicates that the fields NPWG, NPWA, NPDG, NPDA, and
34	 * NOWS are defined for this namespace and should be used by
35	 * the host for I/O optimization.
36	 */
37	id->nsfeat |= 1 << 4;
38	/* NPWG = Namespace Preferred Write Granularity. 0's based */
39	id->npwg = lpp0b;
40	/* NPWA = Namespace Preferred Write Alignment. 0's based */
41	id->npwa = id->npwg;
42	/* NPDG = Namespace Preferred Deallocate Granularity. 0's based */
43	id->npdg = to0based(bdev_discard_granularity(bdev) /
44			    bdev_logical_block_size(bdev));
45	/* NPDG = Namespace Preferred Deallocate Alignment */
46	id->npda = id->npdg;
47	/* NOWS = Namespace Optimal Write Size */
48	id->nows = to0based(bdev_io_opt(bdev) / bdev_logical_block_size(bdev));
49}
50
51void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
52{
53	if (ns->bdev_file) {
54		fput(ns->bdev_file);
55		ns->bdev = NULL;
56		ns->bdev_file = NULL;
57	}
58}
59
60static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns)
61{
62	struct blk_integrity *bi = bdev_get_integrity(ns->bdev);
63
64	if (bi) {
65		ns->metadata_size = bi->tuple_size;
66		if (bi->profile == &t10_pi_type1_crc)
67			ns->pi_type = NVME_NS_DPS_PI_TYPE1;
68		else if (bi->profile == &t10_pi_type3_crc)
69			ns->pi_type = NVME_NS_DPS_PI_TYPE3;
70		else
71			/* Unsupported metadata type */
72			ns->metadata_size = 0;
73	}
74}
75
76int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
77{
78	int ret;
79
80	/*
81	 * When buffered_io namespace attribute is enabled that means user want
82	 * this block device to be used as a file, so block device can take
83	 * an advantage of cache.
84	 */
85	if (ns->buffered_io)
86		return -ENOTBLK;
87
88	ns->bdev_file = bdev_file_open_by_path(ns->device_path,
89				BLK_OPEN_READ | BLK_OPEN_WRITE, NULL, NULL);
90	if (IS_ERR(ns->bdev_file)) {
91		ret = PTR_ERR(ns->bdev_file);
92		if (ret != -ENOTBLK) {
93			pr_err("failed to open block device %s: (%d)\n",
94					ns->device_path, ret);
95		}
96		ns->bdev_file = NULL;
97		return ret;
98	}
99	ns->bdev = file_bdev(ns->bdev_file);
100	ns->size = bdev_nr_bytes(ns->bdev);
101	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
102
103	ns->pi_type = 0;
104	ns->metadata_size = 0;
105	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY_T10))
106		nvmet_bdev_ns_enable_integrity(ns);
107
108	if (bdev_is_zoned(ns->bdev)) {
109		if (!nvmet_bdev_zns_enable(ns)) {
110			nvmet_bdev_ns_disable(ns);
111			return -EINVAL;
112		}
113		ns->csi = NVME_CSI_ZNS;
114	}
115
116	return 0;
117}
118
119void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
120{
121	ns->size = bdev_nr_bytes(ns->bdev);
122}
123
124u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
125{
126	u16 status = NVME_SC_SUCCESS;
127
128	if (likely(blk_sts == BLK_STS_OK))
129		return status;
130	/*
131	 * Right now there exists M : 1 mapping between block layer error
132	 * to the NVMe status code (see nvme_error_status()). For consistency,
133	 * when we reverse map we use most appropriate NVMe Status code from
134	 * the group of the NVMe staus codes used in the nvme_error_status().
135	 */
136	switch (blk_sts) {
137	case BLK_STS_NOSPC:
138		status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
139		req->error_loc = offsetof(struct nvme_rw_command, length);
140		break;
141	case BLK_STS_TARGET:
142		status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
143		req->error_loc = offsetof(struct nvme_rw_command, slba);
144		break;
145	case BLK_STS_NOTSUPP:
146		req->error_loc = offsetof(struct nvme_common_command, opcode);
147		switch (req->cmd->common.opcode) {
148		case nvme_cmd_dsm:
149		case nvme_cmd_write_zeroes:
150			status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
151			break;
152		default:
153			status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
154		}
155		break;
156	case BLK_STS_MEDIUM:
157		status = NVME_SC_ACCESS_DENIED;
158		req->error_loc = offsetof(struct nvme_rw_command, nsid);
159		break;
160	case BLK_STS_IOERR:
161	default:
162		status = NVME_SC_INTERNAL | NVME_SC_DNR;
163		req->error_loc = offsetof(struct nvme_common_command, opcode);
164	}
165
166	switch (req->cmd->common.opcode) {
167	case nvme_cmd_read:
168	case nvme_cmd_write:
169		req->error_slba = le64_to_cpu(req->cmd->rw.slba);
170		break;
171	case nvme_cmd_write_zeroes:
172		req->error_slba =
173			le64_to_cpu(req->cmd->write_zeroes.slba);
174		break;
175	default:
176		req->error_slba = 0;
177	}
178	return status;
179}
180
181static void nvmet_bio_done(struct bio *bio)
182{
183	struct nvmet_req *req = bio->bi_private;
184
185	nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
186	nvmet_req_bio_put(req, bio);
187}
188
189#ifdef CONFIG_BLK_DEV_INTEGRITY
190static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
191				struct sg_mapping_iter *miter)
192{
193	struct blk_integrity *bi;
194	struct bio_integrity_payload *bip;
195	int rc;
196	size_t resid, len;
197
198	bi = bdev_get_integrity(req->ns->bdev);
199	if (unlikely(!bi)) {
200		pr_err("Unable to locate bio_integrity\n");
201		return -ENODEV;
202	}
203
204	bip = bio_integrity_alloc(bio, GFP_NOIO,
205					bio_max_segs(req->metadata_sg_cnt));
206	if (IS_ERR(bip)) {
207		pr_err("Unable to allocate bio_integrity_payload\n");
208		return PTR_ERR(bip);
209	}
210
211	/* virtual start sector must be in integrity interval units */
212	bip_set_seed(bip, bio->bi_iter.bi_sector >>
213		     (bi->interval_exp - SECTOR_SHIFT));
214
215	resid = bio_integrity_bytes(bi, bio_sectors(bio));
216	while (resid > 0 && sg_miter_next(miter)) {
217		len = min_t(size_t, miter->length, resid);
218		rc = bio_integrity_add_page(bio, miter->page, len,
219					    offset_in_page(miter->addr));
220		if (unlikely(rc != len)) {
221			pr_err("bio_integrity_add_page() failed; %d\n", rc);
222			sg_miter_stop(miter);
223			return -ENOMEM;
224		}
225
226		resid -= len;
227		if (len < miter->length)
228			miter->consumed -= miter->length - len;
229	}
230	sg_miter_stop(miter);
231
232	return 0;
233}
234#else
235static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
236				struct sg_mapping_iter *miter)
237{
238	return -EINVAL;
239}
240#endif /* CONFIG_BLK_DEV_INTEGRITY */
241
242static void nvmet_bdev_execute_rw(struct nvmet_req *req)
243{
244	unsigned int sg_cnt = req->sg_cnt;
245	struct bio *bio;
246	struct scatterlist *sg;
247	struct blk_plug plug;
248	sector_t sector;
249	blk_opf_t opf;
250	int i, rc;
251	struct sg_mapping_iter prot_miter;
252	unsigned int iter_flags;
253	unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
254
255	if (!nvmet_check_transfer_len(req, total_len))
256		return;
257
258	if (!req->sg_cnt) {
259		nvmet_req_complete(req, 0);
260		return;
261	}
262
263	if (req->cmd->rw.opcode == nvme_cmd_write) {
264		opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
265		if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
266			opf |= REQ_FUA;
267		iter_flags = SG_MITER_TO_SG;
268	} else {
269		opf = REQ_OP_READ;
270		iter_flags = SG_MITER_FROM_SG;
271	}
272
273	if (is_pci_p2pdma_page(sg_page(req->sg)))
274		opf |= REQ_NOMERGE;
275
276	sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
277
278	if (nvmet_use_inline_bvec(req)) {
279		bio = &req->b.inline_bio;
280		bio_init(bio, req->ns->bdev, req->inline_bvec,
281			 ARRAY_SIZE(req->inline_bvec), opf);
282	} else {
283		bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf,
284				GFP_KERNEL);
285	}
286	bio->bi_iter.bi_sector = sector;
287	bio->bi_private = req;
288	bio->bi_end_io = nvmet_bio_done;
289
290	blk_start_plug(&plug);
291	if (req->metadata_len)
292		sg_miter_start(&prot_miter, req->metadata_sg,
293			       req->metadata_sg_cnt, iter_flags);
294
295	for_each_sg(req->sg, sg, req->sg_cnt, i) {
296		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
297				!= sg->length) {
298			struct bio *prev = bio;
299
300			if (req->metadata_len) {
301				rc = nvmet_bdev_alloc_bip(req, bio,
302							  &prot_miter);
303				if (unlikely(rc)) {
304					bio_io_error(bio);
305					return;
306				}
307			}
308
309			bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt),
310					opf, GFP_KERNEL);
311			bio->bi_iter.bi_sector = sector;
312
313			bio_chain(bio, prev);
314			submit_bio(prev);
315		}
316
317		sector += sg->length >> 9;
318		sg_cnt--;
319	}
320
321	if (req->metadata_len) {
322		rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
323		if (unlikely(rc)) {
324			bio_io_error(bio);
325			return;
326		}
327	}
328
329	submit_bio(bio);
330	blk_finish_plug(&plug);
331}
332
333static void nvmet_bdev_execute_flush(struct nvmet_req *req)
334{
335	struct bio *bio = &req->b.inline_bio;
336
337	if (!bdev_write_cache(req->ns->bdev)) {
338		nvmet_req_complete(req, NVME_SC_SUCCESS);
339		return;
340	}
341
342	if (!nvmet_check_transfer_len(req, 0))
343		return;
344
345	bio_init(bio, req->ns->bdev, req->inline_bvec,
346		 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH);
347	bio->bi_private = req;
348	bio->bi_end_io = nvmet_bio_done;
349
350	submit_bio(bio);
351}
352
353u16 nvmet_bdev_flush(struct nvmet_req *req)
354{
355	if (!bdev_write_cache(req->ns->bdev))
356		return 0;
357
358	if (blkdev_issue_flush(req->ns->bdev))
359		return NVME_SC_INTERNAL | NVME_SC_DNR;
360	return 0;
361}
362
363static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
364		struct nvme_dsm_range *range, struct bio **bio)
365{
366	struct nvmet_ns *ns = req->ns;
367	int ret;
368
369	ret = __blkdev_issue_discard(ns->bdev,
370			nvmet_lba_to_sect(ns, range->slba),
371			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
372			GFP_KERNEL, bio);
373	if (ret && ret != -EOPNOTSUPP) {
374		req->error_slba = le64_to_cpu(range->slba);
375		return errno_to_nvme_status(req, ret);
376	}
377	return NVME_SC_SUCCESS;
378}
379
380static void nvmet_bdev_execute_discard(struct nvmet_req *req)
381{
382	struct nvme_dsm_range range;
383	struct bio *bio = NULL;
384	int i;
385	u16 status;
386
387	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
388		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
389				sizeof(range));
390		if (status)
391			break;
392
393		status = nvmet_bdev_discard_range(req, &range, &bio);
394		if (status)
395			break;
396	}
397
398	if (bio) {
399		bio->bi_private = req;
400		bio->bi_end_io = nvmet_bio_done;
401		if (status)
402			bio_io_error(bio);
403		else
404			submit_bio(bio);
405	} else {
406		nvmet_req_complete(req, status);
407	}
408}
409
410static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
411{
412	if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
413		return;
414
415	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
416	case NVME_DSMGMT_AD:
417		nvmet_bdev_execute_discard(req);
418		return;
419	case NVME_DSMGMT_IDR:
420	case NVME_DSMGMT_IDW:
421	default:
422		/* Not supported yet */
423		nvmet_req_complete(req, 0);
424		return;
425	}
426}
427
428static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
429{
430	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
431	struct bio *bio = NULL;
432	sector_t sector;
433	sector_t nr_sector;
434	int ret;
435
436	if (!nvmet_check_transfer_len(req, 0))
437		return;
438
439	sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
440	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
441		(req->ns->blksize_shift - 9));
442
443	ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
444			GFP_KERNEL, &bio, 0);
445	if (bio) {
446		bio->bi_private = req;
447		bio->bi_end_io = nvmet_bio_done;
448		submit_bio(bio);
449	} else {
450		nvmet_req_complete(req, errno_to_nvme_status(req, ret));
451	}
452}
453
454u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
455{
456	switch (req->cmd->common.opcode) {
457	case nvme_cmd_read:
458	case nvme_cmd_write:
459		req->execute = nvmet_bdev_execute_rw;
460		if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
461			req->metadata_len = nvmet_rw_metadata_len(req);
462		return 0;
463	case nvme_cmd_flush:
464		req->execute = nvmet_bdev_execute_flush;
465		return 0;
466	case nvme_cmd_dsm:
467		req->execute = nvmet_bdev_execute_dsm;
468		return 0;
469	case nvme_cmd_write_zeroes:
470		req->execute = nvmet_bdev_execute_write_zeroes;
471		return 0;
472	default:
473		return nvmet_report_invalid_opcode(req);
474	}
475}
476