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/module.h>
9 #include "nvmet.h"
10 
11 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)
12 {
13 	const struct queue_limits *ql = &bdev_get_queue(bdev)->limits;
14 	/* Number of logical blocks per physical block. */
15 	const u32 lpp = ql->physical_block_size / ql->logical_block_size;
16 	/* Logical blocks per physical block, 0's based. */
17 	const __le16 lpp0b = to0based(lpp);
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(ql->discard_granularity / ql->logical_block_size);
44 	/* NPDG = Namespace Preferred Deallocate Alignment */
45 	id->npda = id->npdg;
46 	/* NOWS = Namespace Optimal Write Size */
47 	id->nows = to0based(ql->io_opt / ql->logical_block_size);
48 }
49 
50 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
51 {
52 	int ret;
53 
54 	ns->bdev = blkdev_get_by_path(ns->device_path,
55 			FMODE_READ | FMODE_WRITE, NULL);
56 	if (IS_ERR(ns->bdev)) {
57 		ret = PTR_ERR(ns->bdev);
58 		if (ret != -ENOTBLK) {
59 			pr_err("failed to open block device %s: (%ld)\n",
60 					ns->device_path, PTR_ERR(ns->bdev));
61 		}
62 		ns->bdev = NULL;
63 		return ret;
64 	}
65 	ns->size = i_size_read(ns->bdev->bd_inode);
66 	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
67 	return 0;
68 }
69 
70 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
71 {
72 	if (ns->bdev) {
73 		blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
74 		ns->bdev = NULL;
75 	}
76 }
77 
78 static u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
79 {
80 	u16 status = NVME_SC_SUCCESS;
81 
82 	if (likely(blk_sts == BLK_STS_OK))
83 		return status;
84 	/*
85 	 * Right now there exists M : 1 mapping between block layer error
86 	 * to the NVMe status code (see nvme_error_status()). For consistency,
87 	 * when we reverse map we use most appropriate NVMe Status code from
88 	 * the group of the NVMe staus codes used in the nvme_error_status().
89 	 */
90 	switch (blk_sts) {
91 	case BLK_STS_NOSPC:
92 		status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
93 		req->error_loc = offsetof(struct nvme_rw_command, length);
94 		break;
95 	case BLK_STS_TARGET:
96 		status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
97 		req->error_loc = offsetof(struct nvme_rw_command, slba);
98 		break;
99 	case BLK_STS_NOTSUPP:
100 		req->error_loc = offsetof(struct nvme_common_command, opcode);
101 		switch (req->cmd->common.opcode) {
102 		case nvme_cmd_dsm:
103 		case nvme_cmd_write_zeroes:
104 			status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
105 			break;
106 		default:
107 			status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
108 		}
109 		break;
110 	case BLK_STS_MEDIUM:
111 		status = NVME_SC_ACCESS_DENIED;
112 		req->error_loc = offsetof(struct nvme_rw_command, nsid);
113 		break;
114 	case BLK_STS_IOERR:
115 		/* fallthru */
116 	default:
117 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
118 		req->error_loc = offsetof(struct nvme_common_command, opcode);
119 	}
120 
121 	switch (req->cmd->common.opcode) {
122 	case nvme_cmd_read:
123 	case nvme_cmd_write:
124 		req->error_slba = le64_to_cpu(req->cmd->rw.slba);
125 		break;
126 	case nvme_cmd_write_zeroes:
127 		req->error_slba =
128 			le64_to_cpu(req->cmd->write_zeroes.slba);
129 		break;
130 	default:
131 		req->error_slba = 0;
132 	}
133 	return status;
134 }
135 
136 static void nvmet_bio_done(struct bio *bio)
137 {
138 	struct nvmet_req *req = bio->bi_private;
139 
140 	nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
141 	if (bio != &req->b.inline_bio)
142 		bio_put(bio);
143 }
144 
145 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
146 {
147 	int sg_cnt = req->sg_cnt;
148 	struct bio *bio;
149 	struct scatterlist *sg;
150 	struct blk_plug plug;
151 	sector_t sector;
152 	int op, i;
153 
154 	if (!nvmet_check_data_len(req, nvmet_rw_len(req)))
155 		return;
156 
157 	if (!req->sg_cnt) {
158 		nvmet_req_complete(req, 0);
159 		return;
160 	}
161 
162 	if (req->cmd->rw.opcode == nvme_cmd_write) {
163 		op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
164 		if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
165 			op |= REQ_FUA;
166 	} else {
167 		op = REQ_OP_READ;
168 	}
169 
170 	if (is_pci_p2pdma_page(sg_page(req->sg)))
171 		op |= REQ_NOMERGE;
172 
173 	sector = le64_to_cpu(req->cmd->rw.slba);
174 	sector <<= (req->ns->blksize_shift - 9);
175 
176 	if (req->transfer_len <= NVMET_MAX_INLINE_DATA_LEN) {
177 		bio = &req->b.inline_bio;
178 		bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
179 	} else {
180 		bio = bio_alloc(GFP_KERNEL, min(sg_cnt, BIO_MAX_PAGES));
181 	}
182 	bio_set_dev(bio, req->ns->bdev);
183 	bio->bi_iter.bi_sector = sector;
184 	bio->bi_private = req;
185 	bio->bi_end_io = nvmet_bio_done;
186 	bio->bi_opf = op;
187 
188 	blk_start_plug(&plug);
189 	for_each_sg(req->sg, sg, req->sg_cnt, i) {
190 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
191 				!= sg->length) {
192 			struct bio *prev = bio;
193 
194 			bio = bio_alloc(GFP_KERNEL, min(sg_cnt, BIO_MAX_PAGES));
195 			bio_set_dev(bio, req->ns->bdev);
196 			bio->bi_iter.bi_sector = sector;
197 			bio->bi_opf = op;
198 
199 			bio_chain(bio, prev);
200 			submit_bio(prev);
201 		}
202 
203 		sector += sg->length >> 9;
204 		sg_cnt--;
205 	}
206 
207 	submit_bio(bio);
208 	blk_finish_plug(&plug);
209 }
210 
211 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
212 {
213 	struct bio *bio = &req->b.inline_bio;
214 
215 	if (!nvmet_check_data_len(req, 0))
216 		return;
217 
218 	bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
219 	bio_set_dev(bio, req->ns->bdev);
220 	bio->bi_private = req;
221 	bio->bi_end_io = nvmet_bio_done;
222 	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
223 
224 	submit_bio(bio);
225 }
226 
227 u16 nvmet_bdev_flush(struct nvmet_req *req)
228 {
229 	if (blkdev_issue_flush(req->ns->bdev, GFP_KERNEL, NULL))
230 		return NVME_SC_INTERNAL | NVME_SC_DNR;
231 	return 0;
232 }
233 
234 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
235 		struct nvme_dsm_range *range, struct bio **bio)
236 {
237 	struct nvmet_ns *ns = req->ns;
238 	int ret;
239 
240 	ret = __blkdev_issue_discard(ns->bdev,
241 			le64_to_cpu(range->slba) << (ns->blksize_shift - 9),
242 			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
243 			GFP_KERNEL, 0, bio);
244 	if (ret && ret != -EOPNOTSUPP) {
245 		req->error_slba = le64_to_cpu(range->slba);
246 		return errno_to_nvme_status(req, ret);
247 	}
248 	return NVME_SC_SUCCESS;
249 }
250 
251 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
252 {
253 	struct nvme_dsm_range range;
254 	struct bio *bio = NULL;
255 	int i;
256 	u16 status;
257 
258 	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
259 		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
260 				sizeof(range));
261 		if (status)
262 			break;
263 
264 		status = nvmet_bdev_discard_range(req, &range, &bio);
265 		if (status)
266 			break;
267 	}
268 
269 	if (bio) {
270 		bio->bi_private = req;
271 		bio->bi_end_io = nvmet_bio_done;
272 		if (status)
273 			bio_io_error(bio);
274 		else
275 			submit_bio(bio);
276 	} else {
277 		nvmet_req_complete(req, status);
278 	}
279 }
280 
281 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
282 {
283 	if (!nvmet_check_data_len(req, nvmet_dsm_len(req)))
284 		return;
285 
286 	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
287 	case NVME_DSMGMT_AD:
288 		nvmet_bdev_execute_discard(req);
289 		return;
290 	case NVME_DSMGMT_IDR:
291 	case NVME_DSMGMT_IDW:
292 	default:
293 		/* Not supported yet */
294 		nvmet_req_complete(req, 0);
295 		return;
296 	}
297 }
298 
299 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
300 {
301 	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
302 	struct bio *bio = NULL;
303 	sector_t sector;
304 	sector_t nr_sector;
305 	int ret;
306 
307 	if (!nvmet_check_data_len(req, 0))
308 		return;
309 
310 	sector = le64_to_cpu(write_zeroes->slba) <<
311 		(req->ns->blksize_shift - 9);
312 	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
313 		(req->ns->blksize_shift - 9));
314 
315 	ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
316 			GFP_KERNEL, &bio, 0);
317 	if (bio) {
318 		bio->bi_private = req;
319 		bio->bi_end_io = nvmet_bio_done;
320 		submit_bio(bio);
321 	} else {
322 		nvmet_req_complete(req, errno_to_nvme_status(req, ret));
323 	}
324 }
325 
326 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
327 {
328 	struct nvme_command *cmd = req->cmd;
329 
330 	switch (cmd->common.opcode) {
331 	case nvme_cmd_read:
332 	case nvme_cmd_write:
333 		req->execute = nvmet_bdev_execute_rw;
334 		return 0;
335 	case nvme_cmd_flush:
336 		req->execute = nvmet_bdev_execute_flush;
337 		return 0;
338 	case nvme_cmd_dsm:
339 		req->execute = nvmet_bdev_execute_dsm;
340 		return 0;
341 	case nvme_cmd_write_zeroes:
342 		req->execute = nvmet_bdev_execute_write_zeroes;
343 		return 0;
344 	default:
345 		pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
346 		       req->sq->qid);
347 		req->error_loc = offsetof(struct nvme_common_command, opcode);
348 		return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
349 	}
350 }
351