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 = &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 } 113 114 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 } 126 127 static u16 nvmet_bdev_discard_range(struct nvmet_ns *ns, 128 struct nvme_dsm_range *range, struct bio **bio) 129 { 130 int ret; 131 132 ret = __blkdev_issue_discard(ns->bdev, 133 le64_to_cpu(range->slba) << (ns->blksize_shift - 9), 134 le32_to_cpu(range->nlb) << (ns->blksize_shift - 9), 135 GFP_KERNEL, 0, bio); 136 if (ret && ret != -EOPNOTSUPP) 137 return NVME_SC_INTERNAL | NVME_SC_DNR; 138 return 0; 139 } 140 141 static void nvmet_bdev_execute_discard(struct nvmet_req *req) 142 { 143 struct nvme_dsm_range range; 144 struct bio *bio = NULL; 145 int i; 146 u16 status; 147 148 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) { 149 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range, 150 sizeof(range)); 151 if (status) 152 break; 153 154 status = nvmet_bdev_discard_range(req->ns, &range, &bio); 155 if (status) 156 break; 157 } 158 159 if (bio) { 160 bio->bi_private = req; 161 bio->bi_end_io = nvmet_bio_done; 162 if (status) { 163 bio->bi_status = BLK_STS_IOERR; 164 bio_endio(bio); 165 } else { 166 submit_bio(bio); 167 } 168 } else { 169 nvmet_req_complete(req, status); 170 } 171 } 172 173 static void nvmet_bdev_execute_dsm(struct nvmet_req *req) 174 { 175 switch (le32_to_cpu(req->cmd->dsm.attributes)) { 176 case NVME_DSMGMT_AD: 177 nvmet_bdev_execute_discard(req); 178 return; 179 case NVME_DSMGMT_IDR: 180 case NVME_DSMGMT_IDW: 181 default: 182 /* Not supported yet */ 183 nvmet_req_complete(req, 0); 184 return; 185 } 186 } 187 188 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req) 189 { 190 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes; 191 struct bio *bio = NULL; 192 u16 status = NVME_SC_SUCCESS; 193 sector_t sector; 194 sector_t nr_sector; 195 196 sector = le64_to_cpu(write_zeroes->slba) << 197 (req->ns->blksize_shift - 9); 198 nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) << 199 (req->ns->blksize_shift - 9)); 200 201 if (__blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector, 202 GFP_KERNEL, &bio, 0)) 203 status = NVME_SC_INTERNAL | NVME_SC_DNR; 204 205 if (bio) { 206 bio->bi_private = req; 207 bio->bi_end_io = nvmet_bio_done; 208 submit_bio(bio); 209 } else { 210 nvmet_req_complete(req, status); 211 } 212 } 213 214 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req) 215 { 216 struct nvme_command *cmd = req->cmd; 217 218 switch (cmd->common.opcode) { 219 case nvme_cmd_read: 220 case nvme_cmd_write: 221 req->execute = nvmet_bdev_execute_rw; 222 req->data_len = nvmet_rw_len(req); 223 return 0; 224 case nvme_cmd_flush: 225 req->execute = nvmet_bdev_execute_flush; 226 req->data_len = 0; 227 return 0; 228 case nvme_cmd_dsm: 229 req->execute = nvmet_bdev_execute_dsm; 230 req->data_len = (le32_to_cpu(cmd->dsm.nr) + 1) * 231 sizeof(struct nvme_dsm_range); 232 return 0; 233 case nvme_cmd_write_zeroes: 234 req->execute = nvmet_bdev_execute_write_zeroes; 235 return 0; 236 default: 237 pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode, 238 req->sq->qid); 239 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR; 240 } 241 } 242