1 /*
2  * NVMe I/O command implementation.
3  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/blkdev.h>
16 #include <linux/module.h>
17 #include "nvmet.h"
18 
nvmet_bdev_ns_enable(struct nvmet_ns * ns)19 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
20 {
21 	int ret;
22 
23 	ns->bdev = blkdev_get_by_path(ns->device_path,
24 			FMODE_READ | FMODE_WRITE, NULL);
25 	if (IS_ERR(ns->bdev)) {
26 		ret = PTR_ERR(ns->bdev);
27 		if (ret != -ENOTBLK) {
28 			pr_err("failed to open block device %s: (%ld)\n",
29 					ns->device_path, PTR_ERR(ns->bdev));
30 		}
31 		ns->bdev = NULL;
32 		return ret;
33 	}
34 	ns->size = i_size_read(ns->bdev->bd_inode);
35 	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
36 	return 0;
37 }
38 
nvmet_bdev_ns_disable(struct nvmet_ns * ns)39 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
40 {
41 	if (ns->bdev) {
42 		blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
43 		ns->bdev = NULL;
44 	}
45 }
46 
nvmet_bio_done(struct bio * bio)47 static void nvmet_bio_done(struct bio *bio)
48 {
49 	struct nvmet_req *req = bio->bi_private;
50 
51 	nvmet_req_complete(req,
52 		bio->bi_status ? NVME_SC_INTERNAL | NVME_SC_DNR : 0);
53 
54 	if (bio != &req->b.inline_bio)
55 		bio_put(bio);
56 }
57 
nvmet_bdev_execute_rw(struct nvmet_req * req)58 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
59 {
60 	int sg_cnt = req->sg_cnt;
61 	struct bio *bio = &req->b.inline_bio;
62 	struct scatterlist *sg;
63 	sector_t sector;
64 	blk_qc_t cookie;
65 	int op, op_flags = 0, i;
66 
67 	if (!req->sg_cnt) {
68 		nvmet_req_complete(req, 0);
69 		return;
70 	}
71 
72 	if (req->cmd->rw.opcode == nvme_cmd_write) {
73 		op = REQ_OP_WRITE;
74 		op_flags = REQ_SYNC | REQ_IDLE;
75 		if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
76 			op_flags |= REQ_FUA;
77 	} else {
78 		op = REQ_OP_READ;
79 	}
80 
81 	sector = le64_to_cpu(req->cmd->rw.slba);
82 	sector <<= (req->ns->blksize_shift - 9);
83 
84 	bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
85 	bio_set_dev(bio, req->ns->bdev);
86 	bio->bi_iter.bi_sector = sector;
87 	bio->bi_private = req;
88 	bio->bi_end_io = nvmet_bio_done;
89 	bio_set_op_attrs(bio, op, op_flags);
90 
91 	for_each_sg(req->sg, sg, req->sg_cnt, i) {
92 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
93 				!= sg->length) {
94 			struct bio *prev = bio;
95 
96 			bio = bio_alloc(GFP_KERNEL, min(sg_cnt, BIO_MAX_PAGES));
97 			bio_set_dev(bio, req->ns->bdev);
98 			bio->bi_iter.bi_sector = sector;
99 			bio_set_op_attrs(bio, op, op_flags);
100 
101 			bio_chain(bio, prev);
102 			submit_bio(prev);
103 		}
104 
105 		sector += sg->length >> 9;
106 		sg_cnt--;
107 	}
108 
109 	cookie = submit_bio(bio);
110 
111 	blk_poll(bdev_get_queue(req->ns->bdev), cookie);
112 }
113 
nvmet_bdev_execute_flush(struct nvmet_req * req)114 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
115 {
116 	struct bio *bio = &req->b.inline_bio;
117 
118 	bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
119 	bio_set_dev(bio, req->ns->bdev);
120 	bio->bi_private = req;
121 	bio->bi_end_io = nvmet_bio_done;
122 	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
123 
124 	submit_bio(bio);
125 }
126 
nvmet_bdev_flush(struct nvmet_req * req)127 u16 nvmet_bdev_flush(struct nvmet_req *req)
128 {
129 	if (blkdev_issue_flush(req->ns->bdev, GFP_KERNEL, NULL))
130 		return NVME_SC_INTERNAL | NVME_SC_DNR;
131 	return 0;
132 }
133 
nvmet_bdev_discard_range(struct nvmet_ns * ns,struct nvme_dsm_range * range,struct bio ** bio)134 static u16 nvmet_bdev_discard_range(struct nvmet_ns *ns,
135 		struct nvme_dsm_range *range, struct bio **bio)
136 {
137 	int ret;
138 
139 	ret = __blkdev_issue_discard(ns->bdev,
140 			le64_to_cpu(range->slba) << (ns->blksize_shift - 9),
141 			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
142 			GFP_KERNEL, 0, bio);
143 	if (ret && ret != -EOPNOTSUPP)
144 		return NVME_SC_INTERNAL | NVME_SC_DNR;
145 	return 0;
146 }
147 
nvmet_bdev_execute_discard(struct nvmet_req * req)148 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
149 {
150 	struct nvme_dsm_range range;
151 	struct bio *bio = NULL;
152 	int i;
153 	u16 status;
154 
155 	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
156 		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
157 				sizeof(range));
158 		if (status)
159 			break;
160 
161 		status = nvmet_bdev_discard_range(req->ns, &range, &bio);
162 		if (status)
163 			break;
164 	}
165 
166 	if (bio) {
167 		bio->bi_private = req;
168 		bio->bi_end_io = nvmet_bio_done;
169 		if (status) {
170 			bio->bi_status = BLK_STS_IOERR;
171 			bio_endio(bio);
172 		} else {
173 			submit_bio(bio);
174 		}
175 	} else {
176 		nvmet_req_complete(req, status);
177 	}
178 }
179 
nvmet_bdev_execute_dsm(struct nvmet_req * req)180 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
181 {
182 	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
183 	case NVME_DSMGMT_AD:
184 		nvmet_bdev_execute_discard(req);
185 		return;
186 	case NVME_DSMGMT_IDR:
187 	case NVME_DSMGMT_IDW:
188 	default:
189 		/* Not supported yet */
190 		nvmet_req_complete(req, 0);
191 		return;
192 	}
193 }
194 
nvmet_bdev_execute_write_zeroes(struct nvmet_req * req)195 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
196 {
197 	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
198 	struct bio *bio = NULL;
199 	u16 status = NVME_SC_SUCCESS;
200 	sector_t sector;
201 	sector_t nr_sector;
202 
203 	sector = le64_to_cpu(write_zeroes->slba) <<
204 		(req->ns->blksize_shift - 9);
205 	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
206 		(req->ns->blksize_shift - 9));
207 
208 	if (__blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
209 				GFP_KERNEL, &bio, 0))
210 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
211 
212 	if (bio) {
213 		bio->bi_private = req;
214 		bio->bi_end_io = nvmet_bio_done;
215 		submit_bio(bio);
216 	} else {
217 		nvmet_req_complete(req, status);
218 	}
219 }
220 
nvmet_bdev_parse_io_cmd(struct nvmet_req * req)221 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
222 {
223 	struct nvme_command *cmd = req->cmd;
224 
225 	switch (cmd->common.opcode) {
226 	case nvme_cmd_read:
227 	case nvme_cmd_write:
228 		req->execute = nvmet_bdev_execute_rw;
229 		req->data_len = nvmet_rw_len(req);
230 		return 0;
231 	case nvme_cmd_flush:
232 		req->execute = nvmet_bdev_execute_flush;
233 		req->data_len = 0;
234 		return 0;
235 	case nvme_cmd_dsm:
236 		req->execute = nvmet_bdev_execute_dsm;
237 		req->data_len = (le32_to_cpu(cmd->dsm.nr) + 1) *
238 			sizeof(struct nvme_dsm_range);
239 		return 0;
240 	case nvme_cmd_write_zeroes:
241 		req->execute = nvmet_bdev_execute_write_zeroes;
242 		return 0;
243 	default:
244 		pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
245 		       req->sq->qid);
246 		return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
247 	}
248 }
249