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 13 void 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 = to0based(bdev_io_min(bdev) / bdev_logical_block_size(bdev)); 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 /* Set WZDS and DRB if device supports unmapped write zeroes */ 51 if (bdev_write_zeroes_unmap_sectors(bdev)) 52 id->dlfeat = (1 << 3) | 0x1; 53 } 54 55 void nvmet_bdev_ns_disable(struct nvmet_ns *ns) 56 { 57 if (ns->bdev_file) { 58 fput(ns->bdev_file); 59 ns->bdev = NULL; 60 ns->bdev_file = NULL; 61 } 62 } 63 64 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns) 65 { 66 struct blk_integrity *bi = bdev_get_integrity(ns->bdev); 67 68 if (!bi) 69 return; 70 71 if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC) { 72 ns->metadata_size = bi->metadata_size; 73 if (bi->flags & BLK_INTEGRITY_REF_TAG) 74 ns->pi_type = NVME_NS_DPS_PI_TYPE1; 75 else 76 ns->pi_type = NVME_NS_DPS_PI_TYPE3; 77 } else { 78 ns->metadata_size = 0; 79 } 80 } 81 82 int nvmet_bdev_ns_enable(struct nvmet_ns *ns) 83 { 84 int ret; 85 86 /* 87 * When buffered_io namespace attribute is enabled that means user want 88 * this block device to be used as a file, so block device can take 89 * an advantage of cache. 90 */ 91 if (ns->buffered_io) 92 return -ENOTBLK; 93 94 ns->bdev_file = bdev_file_open_by_path(ns->device_path, 95 BLK_OPEN_READ | BLK_OPEN_WRITE, NULL, NULL); 96 if (IS_ERR(ns->bdev_file)) { 97 ret = PTR_ERR(ns->bdev_file); 98 if (ret != -ENOTBLK) { 99 pr_err("failed to open block device %s: (%d)\n", 100 ns->device_path, ret); 101 } 102 ns->bdev_file = NULL; 103 return ret; 104 } 105 ns->bdev = file_bdev(ns->bdev_file); 106 ns->size = bdev_nr_bytes(ns->bdev); 107 ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev)); 108 109 ns->pi_type = 0; 110 ns->metadata_size = 0; 111 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) 112 nvmet_bdev_ns_enable_integrity(ns); 113 114 if (bdev_is_zoned(ns->bdev)) { 115 if (!nvmet_bdev_zns_enable(ns)) { 116 nvmet_bdev_ns_disable(ns); 117 return -EINVAL; 118 } 119 ns->csi = NVME_CSI_ZNS; 120 } 121 122 return 0; 123 } 124 125 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns) 126 { 127 ns->size = bdev_nr_bytes(ns->bdev); 128 } 129 130 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts) 131 { 132 u16 status = NVME_SC_SUCCESS; 133 134 if (likely(blk_sts == BLK_STS_OK)) 135 return status; 136 /* 137 * Right now there exists M : 1 mapping between block layer error 138 * to the NVMe status code (see nvme_error_status()). For consistency, 139 * when we reverse map we use most appropriate NVMe Status code from 140 * the group of the NVMe status codes used in the nvme_error_status(). 141 */ 142 switch (blk_sts) { 143 case BLK_STS_NOSPC: 144 status = NVME_SC_CAP_EXCEEDED | NVME_STATUS_DNR; 145 req->error_loc = offsetof(struct nvme_rw_command, length); 146 break; 147 case BLK_STS_TARGET: 148 status = NVME_SC_LBA_RANGE | NVME_STATUS_DNR; 149 req->error_loc = offsetof(struct nvme_rw_command, slba); 150 break; 151 case BLK_STS_NOTSUPP: 152 status = NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR; 153 req->error_loc = offsetof(struct nvme_common_command, opcode); 154 break; 155 case BLK_STS_MEDIUM: 156 status = NVME_SC_ACCESS_DENIED; 157 req->error_loc = offsetof(struct nvme_rw_command, nsid); 158 break; 159 case BLK_STS_IOERR: 160 default: 161 status = NVME_SC_INTERNAL | NVME_STATUS_DNR; 162 req->error_loc = offsetof(struct nvme_common_command, opcode); 163 } 164 165 switch (req->cmd->common.opcode) { 166 case nvme_cmd_read: 167 case nvme_cmd_write: 168 req->error_slba = le64_to_cpu(req->cmd->rw.slba); 169 break; 170 case nvme_cmd_write_zeroes: 171 req->error_slba = 172 le64_to_cpu(req->cmd->write_zeroes.slba); 173 break; 174 default: 175 req->error_slba = 0; 176 } 177 return status; 178 } 179 180 static void nvmet_bio_done(struct bio *bio) 181 { 182 struct nvmet_req *req = bio->bi_private; 183 184 nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status)); 185 nvmet_req_bio_put(req, bio); 186 } 187 188 #ifdef CONFIG_BLK_DEV_INTEGRITY 189 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio, 190 struct sg_mapping_iter *miter) 191 { 192 struct blk_integrity *bi; 193 struct bio_integrity_payload *bip; 194 int rc; 195 size_t resid, len; 196 197 bi = bdev_get_integrity(req->ns->bdev); 198 if (unlikely(!bi)) { 199 pr_err("Unable to locate bio_integrity\n"); 200 return -ENODEV; 201 } 202 203 bip = bio_integrity_alloc(bio, GFP_NOIO, 204 bio_max_segs(req->metadata_sg_cnt)); 205 if (IS_ERR(bip)) { 206 pr_err("Unable to allocate bio_integrity_payload\n"); 207 return PTR_ERR(bip); 208 } 209 210 /* virtual start sector must be in integrity interval units */ 211 bip_set_seed(bip, bio->bi_iter.bi_sector >> 212 (bi->interval_exp - SECTOR_SHIFT)); 213 214 resid = bio_integrity_bytes(bi, bio_sectors(bio)); 215 while (resid > 0 && sg_miter_next(miter)) { 216 len = min_t(size_t, miter->length, resid); 217 rc = bio_integrity_add_page(bio, miter->page, len, 218 offset_in_page(miter->addr)); 219 if (unlikely(rc != len)) { 220 pr_err("bio_integrity_add_page() failed; %d\n", rc); 221 sg_miter_stop(miter); 222 return -ENOMEM; 223 } 224 225 resid -= len; 226 if (len < miter->length) 227 miter->consumed -= miter->length - len; 228 } 229 sg_miter_stop(miter); 230 231 return 0; 232 } 233 #else 234 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio, 235 struct sg_mapping_iter *miter) 236 { 237 return -EINVAL; 238 } 239 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 240 241 static void nvmet_bdev_execute_rw(struct nvmet_req *req) 242 { 243 unsigned int sg_cnt = req->sg_cnt; 244 struct bio *bio; 245 struct scatterlist *sg; 246 struct blk_plug plug; 247 sector_t sector; 248 blk_opf_t opf; 249 int i, rc; 250 struct sg_mapping_iter prot_miter; 251 unsigned int iter_flags; 252 unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len; 253 254 if (!nvmet_check_transfer_len(req, total_len)) 255 return; 256 257 if (!req->sg_cnt) { 258 nvmet_req_complete(req, 0); 259 return; 260 } 261 262 if (req->cmd->rw.opcode == nvme_cmd_write) { 263 opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; 264 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA)) 265 opf |= REQ_FUA; 266 iter_flags = SG_MITER_TO_SG; 267 } else { 268 opf = REQ_OP_READ; 269 iter_flags = SG_MITER_FROM_SG; 270 } 271 272 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_LR)) 273 opf |= REQ_FAILFAST_DEV; 274 275 if (is_pci_p2pdma_page(sg_page(req->sg))) 276 opf |= REQ_NOMERGE; 277 278 sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba); 279 280 if (nvmet_use_inline_bvec(req)) { 281 bio = &req->b.inline_bio; 282 bio_init(bio, req->ns->bdev, req->inline_bvec, 283 ARRAY_SIZE(req->inline_bvec), opf); 284 } else { 285 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf, 286 GFP_KERNEL); 287 } 288 bio->bi_iter.bi_sector = sector; 289 bio->bi_private = req; 290 bio->bi_end_io = nvmet_bio_done; 291 292 blk_start_plug(&plug); 293 if (req->metadata_len) 294 sg_miter_start(&prot_miter, req->metadata_sg, 295 req->metadata_sg_cnt, iter_flags); 296 297 for_each_sg(req->sg, sg, req->sg_cnt, i) { 298 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset) 299 != sg->length) { 300 struct bio *prev = bio; 301 302 if (req->metadata_len) { 303 rc = nvmet_bdev_alloc_bip(req, bio, 304 &prot_miter); 305 if (unlikely(rc)) { 306 bio_io_error(bio); 307 return; 308 } 309 } 310 311 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), 312 opf, GFP_KERNEL); 313 bio->bi_iter.bi_sector = sector; 314 315 bio_chain(bio, prev); 316 submit_bio(prev); 317 } 318 319 sector += sg->length >> 9; 320 sg_cnt--; 321 } 322 323 if (req->metadata_len) { 324 rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter); 325 if (unlikely(rc)) { 326 bio_io_error(bio); 327 return; 328 } 329 } 330 331 submit_bio(bio); 332 blk_finish_plug(&plug); 333 } 334 335 static void nvmet_bdev_execute_flush(struct nvmet_req *req) 336 { 337 struct bio *bio = &req->b.inline_bio; 338 339 if (!bdev_write_cache(req->ns->bdev)) { 340 nvmet_req_complete(req, NVME_SC_SUCCESS); 341 return; 342 } 343 344 if (!nvmet_check_transfer_len(req, 0)) 345 return; 346 347 bio_init(bio, req->ns->bdev, req->inline_bvec, 348 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH); 349 bio->bi_private = req; 350 bio->bi_end_io = nvmet_bio_done; 351 352 submit_bio(bio); 353 } 354 355 u16 nvmet_bdev_flush(struct nvmet_req *req) 356 { 357 if (!bdev_write_cache(req->ns->bdev)) 358 return 0; 359 360 if (blkdev_issue_flush(req->ns->bdev)) 361 return NVME_SC_INTERNAL | NVME_STATUS_DNR; 362 return 0; 363 } 364 365 static u16 nvmet_bdev_discard_range(struct nvmet_req *req, 366 struct nvme_dsm_range *range, struct bio **bio) 367 { 368 struct nvmet_ns *ns = req->ns; 369 int ret; 370 371 ret = __blkdev_issue_discard(ns->bdev, 372 nvmet_lba_to_sect(ns, range->slba), 373 le32_to_cpu(range->nlb) << (ns->blksize_shift - 9), 374 GFP_KERNEL, bio); 375 if (ret && ret != -EOPNOTSUPP) { 376 req->error_slba = le64_to_cpu(range->slba); 377 return errno_to_nvme_status(req, ret); 378 } 379 return NVME_SC_SUCCESS; 380 } 381 382 static void nvmet_bdev_execute_discard(struct nvmet_req *req) 383 { 384 struct nvme_dsm_range range; 385 struct bio *bio = NULL; 386 int i; 387 u16 status; 388 389 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) { 390 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range, 391 sizeof(range)); 392 if (status) 393 break; 394 395 status = nvmet_bdev_discard_range(req, &range, &bio); 396 if (status) 397 break; 398 } 399 400 if (bio) { 401 bio->bi_private = req; 402 bio->bi_end_io = nvmet_bio_done; 403 if (status) 404 bio_io_error(bio); 405 else 406 submit_bio(bio); 407 } else { 408 nvmet_req_complete(req, status); 409 } 410 } 411 412 static void nvmet_bdev_execute_dsm(struct nvmet_req *req) 413 { 414 if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req))) 415 return; 416 417 switch (le32_to_cpu(req->cmd->dsm.attributes)) { 418 case NVME_DSMGMT_AD: 419 nvmet_bdev_execute_discard(req); 420 return; 421 case NVME_DSMGMT_IDR: 422 case NVME_DSMGMT_IDW: 423 default: 424 /* Not supported yet */ 425 nvmet_req_complete(req, 0); 426 return; 427 } 428 } 429 430 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req) 431 { 432 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes; 433 struct bio *bio = NULL; 434 sector_t sector; 435 sector_t nr_sector; 436 int ret; 437 438 if (!nvmet_check_transfer_len(req, 0)) 439 return; 440 441 sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba); 442 nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) << 443 (req->ns->blksize_shift - 9)); 444 445 ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector, 446 GFP_KERNEL, &bio, 0); 447 if (bio) { 448 bio->bi_private = req; 449 bio->bi_end_io = nvmet_bio_done; 450 submit_bio(bio); 451 } else { 452 nvmet_req_complete(req, errno_to_nvme_status(req, ret)); 453 } 454 } 455 456 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req) 457 { 458 switch (req->cmd->common.opcode) { 459 case nvme_cmd_read: 460 case nvme_cmd_write: 461 req->execute = nvmet_bdev_execute_rw; 462 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns)) 463 req->metadata_len = nvmet_rw_metadata_len(req); 464 return 0; 465 case nvme_cmd_flush: 466 req->execute = nvmet_bdev_execute_flush; 467 return 0; 468 case nvme_cmd_dsm: 469 req->execute = nvmet_bdev_execute_dsm; 470 return 0; 471 case nvme_cmd_write_zeroes: 472 req->execute = nvmet_bdev_execute_write_zeroes; 473 return 0; 474 default: 475 return nvmet_report_invalid_opcode(req); 476 } 477 } 478