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 
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 
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 
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 
58 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
59 {
60 	int sg_cnt = req->sg_cnt;
61 	struct bio *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 	if (is_pci_p2pdma_page(sg_page(req->sg)))
82 		op_flags |= REQ_NOMERGE;
83 
84 	sector = le64_to_cpu(req->cmd->rw.slba);
85 	sector <<= (req->ns->blksize_shift - 9);
86 
87 	if (req->data_len <= NVMET_MAX_INLINE_DATA_LEN) {
88 		bio = &req->b.inline_bio;
89 		bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
90 	} else {
91 		bio = bio_alloc(GFP_KERNEL, min(sg_cnt, BIO_MAX_PAGES));
92 	}
93 	bio_set_dev(bio, req->ns->bdev);
94 	bio->bi_iter.bi_sector = sector;
95 	bio->bi_private = req;
96 	bio->bi_end_io = nvmet_bio_done;
97 	bio_set_op_attrs(bio, op, op_flags);
98 
99 	for_each_sg(req->sg, sg, req->sg_cnt, i) {
100 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
101 				!= sg->length) {
102 			struct bio *prev = bio;
103 
104 			bio = bio_alloc(GFP_KERNEL, min(sg_cnt, BIO_MAX_PAGES));
105 			bio_set_dev(bio, req->ns->bdev);
106 			bio->bi_iter.bi_sector = sector;
107 			bio_set_op_attrs(bio, op, op_flags);
108 
109 			bio_chain(bio, prev);
110 			submit_bio(prev);
111 		}
112 
113 		sector += sg->length >> 9;
114 		sg_cnt--;
115 	}
116 
117 	cookie = submit_bio(bio);
118 
119 	blk_poll(bdev_get_queue(req->ns->bdev), cookie);
120 }
121 
122 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
123 {
124 	struct bio *bio = &req->b.inline_bio;
125 
126 	bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
127 	bio_set_dev(bio, req->ns->bdev);
128 	bio->bi_private = req;
129 	bio->bi_end_io = nvmet_bio_done;
130 	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
131 
132 	submit_bio(bio);
133 }
134 
135 u16 nvmet_bdev_flush(struct nvmet_req *req)
136 {
137 	if (blkdev_issue_flush(req->ns->bdev, GFP_KERNEL, NULL))
138 		return NVME_SC_INTERNAL | NVME_SC_DNR;
139 	return 0;
140 }
141 
142 static u16 nvmet_bdev_discard_range(struct nvmet_ns *ns,
143 		struct nvme_dsm_range *range, struct bio **bio)
144 {
145 	int ret;
146 
147 	ret = __blkdev_issue_discard(ns->bdev,
148 			le64_to_cpu(range->slba) << (ns->blksize_shift - 9),
149 			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
150 			GFP_KERNEL, 0, bio);
151 	if (ret && ret != -EOPNOTSUPP)
152 		return NVME_SC_INTERNAL | NVME_SC_DNR;
153 	return 0;
154 }
155 
156 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
157 {
158 	struct nvme_dsm_range range;
159 	struct bio *bio = NULL;
160 	int i;
161 	u16 status;
162 
163 	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
164 		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
165 				sizeof(range));
166 		if (status)
167 			break;
168 
169 		status = nvmet_bdev_discard_range(req->ns, &range, &bio);
170 		if (status)
171 			break;
172 	}
173 
174 	if (bio) {
175 		bio->bi_private = req;
176 		bio->bi_end_io = nvmet_bio_done;
177 		if (status) {
178 			bio->bi_status = BLK_STS_IOERR;
179 			bio_endio(bio);
180 		} else {
181 			submit_bio(bio);
182 		}
183 	} else {
184 		nvmet_req_complete(req, status);
185 	}
186 }
187 
188 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
189 {
190 	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
191 	case NVME_DSMGMT_AD:
192 		nvmet_bdev_execute_discard(req);
193 		return;
194 	case NVME_DSMGMT_IDR:
195 	case NVME_DSMGMT_IDW:
196 	default:
197 		/* Not supported yet */
198 		nvmet_req_complete(req, 0);
199 		return;
200 	}
201 }
202 
203 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
204 {
205 	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
206 	struct bio *bio = NULL;
207 	u16 status = NVME_SC_SUCCESS;
208 	sector_t sector;
209 	sector_t nr_sector;
210 
211 	sector = le64_to_cpu(write_zeroes->slba) <<
212 		(req->ns->blksize_shift - 9);
213 	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
214 		(req->ns->blksize_shift - 9));
215 
216 	if (__blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
217 				GFP_KERNEL, &bio, 0))
218 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
219 
220 	if (bio) {
221 		bio->bi_private = req;
222 		bio->bi_end_io = nvmet_bio_done;
223 		submit_bio(bio);
224 	} else {
225 		nvmet_req_complete(req, status);
226 	}
227 }
228 
229 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
230 {
231 	struct nvme_command *cmd = req->cmd;
232 
233 	switch (cmd->common.opcode) {
234 	case nvme_cmd_read:
235 	case nvme_cmd_write:
236 		req->execute = nvmet_bdev_execute_rw;
237 		req->data_len = nvmet_rw_len(req);
238 		return 0;
239 	case nvme_cmd_flush:
240 		req->execute = nvmet_bdev_execute_flush;
241 		req->data_len = 0;
242 		return 0;
243 	case nvme_cmd_dsm:
244 		req->execute = nvmet_bdev_execute_dsm;
245 		req->data_len = (le32_to_cpu(cmd->dsm.nr) + 1) *
246 			sizeof(struct nvme_dsm_range);
247 		return 0;
248 	case nvme_cmd_write_zeroes:
249 		req->execute = nvmet_bdev_execute_write_zeroes;
250 		return 0;
251 	default:
252 		pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
253 		       req->sq->qid);
254 		return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
255 	}
256 }
257