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 51 void nvmet_bdev_ns_disable(struct nvmet_ns *ns) 52 { 53 if (ns->bdev_file) { 54 fput(ns->bdev_file); 55 ns->bdev = NULL; 56 ns->bdev_file = NULL; 57 } 58 } 59 60 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns) 61 { 62 struct blk_integrity *bi = bdev_get_integrity(ns->bdev); 63 64 if (!bi) 65 return; 66 67 if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC) { 68 ns->metadata_size = bi->tuple_size; 69 if (bi->flags & BLK_INTEGRITY_REF_TAG) 70 ns->pi_type = NVME_NS_DPS_PI_TYPE1; 71 else 72 ns->pi_type = NVME_NS_DPS_PI_TYPE3; 73 } else { 74 ns->metadata_size = 0; 75 } 76 } 77 78 int nvmet_bdev_ns_enable(struct nvmet_ns *ns) 79 { 80 int ret; 81 82 /* 83 * When buffered_io namespace attribute is enabled that means user want 84 * this block device to be used as a file, so block device can take 85 * an advantage of cache. 86 */ 87 if (ns->buffered_io) 88 return -ENOTBLK; 89 90 ns->bdev_file = bdev_file_open_by_path(ns->device_path, 91 BLK_OPEN_READ | BLK_OPEN_WRITE, NULL, NULL); 92 if (IS_ERR(ns->bdev_file)) { 93 ret = PTR_ERR(ns->bdev_file); 94 if (ret != -ENOTBLK) { 95 pr_err("failed to open block device %s: (%d)\n", 96 ns->device_path, ret); 97 } 98 ns->bdev_file = NULL; 99 return ret; 100 } 101 ns->bdev = file_bdev(ns->bdev_file); 102 ns->size = bdev_nr_bytes(ns->bdev); 103 ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev)); 104 105 ns->pi_type = 0; 106 ns->metadata_size = 0; 107 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) 108 nvmet_bdev_ns_enable_integrity(ns); 109 110 if (bdev_is_zoned(ns->bdev)) { 111 if (!nvmet_bdev_zns_enable(ns)) { 112 nvmet_bdev_ns_disable(ns); 113 return -EINVAL; 114 } 115 ns->csi = NVME_CSI_ZNS; 116 } 117 118 return 0; 119 } 120 121 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns) 122 { 123 ns->size = bdev_nr_bytes(ns->bdev); 124 } 125 126 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts) 127 { 128 u16 status = NVME_SC_SUCCESS; 129 130 if (likely(blk_sts == BLK_STS_OK)) 131 return status; 132 /* 133 * Right now there exists M : 1 mapping between block layer error 134 * to the NVMe status code (see nvme_error_status()). For consistency, 135 * when we reverse map we use most appropriate NVMe Status code from 136 * the group of the NVMe staus codes used in the nvme_error_status(). 137 */ 138 switch (blk_sts) { 139 case BLK_STS_NOSPC: 140 status = NVME_SC_CAP_EXCEEDED | NVME_STATUS_DNR; 141 req->error_loc = offsetof(struct nvme_rw_command, length); 142 break; 143 case BLK_STS_TARGET: 144 status = NVME_SC_LBA_RANGE | NVME_STATUS_DNR; 145 req->error_loc = offsetof(struct nvme_rw_command, slba); 146 break; 147 case BLK_STS_NOTSUPP: 148 req->error_loc = offsetof(struct nvme_common_command, opcode); 149 switch (req->cmd->common.opcode) { 150 case nvme_cmd_dsm: 151 case nvme_cmd_write_zeroes: 152 status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_STATUS_DNR; 153 break; 154 default: 155 status = NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR; 156 } 157 break; 158 case BLK_STS_MEDIUM: 159 status = NVME_SC_ACCESS_DENIED; 160 req->error_loc = offsetof(struct nvme_rw_command, nsid); 161 break; 162 case BLK_STS_IOERR: 163 default: 164 status = NVME_SC_INTERNAL | NVME_STATUS_DNR; 165 req->error_loc = offsetof(struct nvme_common_command, opcode); 166 } 167 168 switch (req->cmd->common.opcode) { 169 case nvme_cmd_read: 170 case nvme_cmd_write: 171 req->error_slba = le64_to_cpu(req->cmd->rw.slba); 172 break; 173 case nvme_cmd_write_zeroes: 174 req->error_slba = 175 le64_to_cpu(req->cmd->write_zeroes.slba); 176 break; 177 default: 178 req->error_slba = 0; 179 } 180 return status; 181 } 182 183 static void nvmet_bio_done(struct bio *bio) 184 { 185 struct nvmet_req *req = bio->bi_private; 186 187 nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status)); 188 nvmet_req_bio_put(req, bio); 189 } 190 191 #ifdef CONFIG_BLK_DEV_INTEGRITY 192 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio, 193 struct sg_mapping_iter *miter) 194 { 195 struct blk_integrity *bi; 196 struct bio_integrity_payload *bip; 197 int rc; 198 size_t resid, len; 199 200 bi = bdev_get_integrity(req->ns->bdev); 201 if (unlikely(!bi)) { 202 pr_err("Unable to locate bio_integrity\n"); 203 return -ENODEV; 204 } 205 206 bip = bio_integrity_alloc(bio, GFP_NOIO, 207 bio_max_segs(req->metadata_sg_cnt)); 208 if (IS_ERR(bip)) { 209 pr_err("Unable to allocate bio_integrity_payload\n"); 210 return PTR_ERR(bip); 211 } 212 213 /* virtual start sector must be in integrity interval units */ 214 bip_set_seed(bip, bio->bi_iter.bi_sector >> 215 (bi->interval_exp - SECTOR_SHIFT)); 216 217 resid = bio_integrity_bytes(bi, bio_sectors(bio)); 218 while (resid > 0 && sg_miter_next(miter)) { 219 len = min_t(size_t, miter->length, resid); 220 rc = bio_integrity_add_page(bio, miter->page, len, 221 offset_in_page(miter->addr)); 222 if (unlikely(rc != len)) { 223 pr_err("bio_integrity_add_page() failed; %d\n", rc); 224 sg_miter_stop(miter); 225 return -ENOMEM; 226 } 227 228 resid -= len; 229 if (len < miter->length) 230 miter->consumed -= miter->length - len; 231 } 232 sg_miter_stop(miter); 233 234 return 0; 235 } 236 #else 237 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio, 238 struct sg_mapping_iter *miter) 239 { 240 return -EINVAL; 241 } 242 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 243 244 static void nvmet_bdev_execute_rw(struct nvmet_req *req) 245 { 246 unsigned int sg_cnt = req->sg_cnt; 247 struct bio *bio; 248 struct scatterlist *sg; 249 struct blk_plug plug; 250 sector_t sector; 251 blk_opf_t opf; 252 int i, rc; 253 struct sg_mapping_iter prot_miter; 254 unsigned int iter_flags; 255 unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len; 256 257 if (!nvmet_check_transfer_len(req, total_len)) 258 return; 259 260 if (!req->sg_cnt) { 261 nvmet_req_complete(req, 0); 262 return; 263 } 264 265 if (req->cmd->rw.opcode == nvme_cmd_write) { 266 opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; 267 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA)) 268 opf |= REQ_FUA; 269 iter_flags = SG_MITER_TO_SG; 270 } else { 271 opf = REQ_OP_READ; 272 iter_flags = SG_MITER_FROM_SG; 273 } 274 275 if (req->cmd->rw.control & NVME_RW_LR) 276 opf |= REQ_FAILFAST_DEV; 277 278 if (is_pci_p2pdma_page(sg_page(req->sg))) 279 opf |= REQ_NOMERGE; 280 281 sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba); 282 283 if (nvmet_use_inline_bvec(req)) { 284 bio = &req->b.inline_bio; 285 bio_init(bio, req->ns->bdev, req->inline_bvec, 286 ARRAY_SIZE(req->inline_bvec), opf); 287 } else { 288 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf, 289 GFP_KERNEL); 290 } 291 bio->bi_iter.bi_sector = sector; 292 bio->bi_private = req; 293 bio->bi_end_io = nvmet_bio_done; 294 295 blk_start_plug(&plug); 296 if (req->metadata_len) 297 sg_miter_start(&prot_miter, req->metadata_sg, 298 req->metadata_sg_cnt, iter_flags); 299 300 for_each_sg(req->sg, sg, req->sg_cnt, i) { 301 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset) 302 != sg->length) { 303 struct bio *prev = bio; 304 305 if (req->metadata_len) { 306 rc = nvmet_bdev_alloc_bip(req, bio, 307 &prot_miter); 308 if (unlikely(rc)) { 309 bio_io_error(bio); 310 return; 311 } 312 } 313 314 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), 315 opf, GFP_KERNEL); 316 bio->bi_iter.bi_sector = sector; 317 318 bio_chain(bio, prev); 319 submit_bio(prev); 320 } 321 322 sector += sg->length >> 9; 323 sg_cnt--; 324 } 325 326 if (req->metadata_len) { 327 rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter); 328 if (unlikely(rc)) { 329 bio_io_error(bio); 330 return; 331 } 332 } 333 334 submit_bio(bio); 335 blk_finish_plug(&plug); 336 } 337 338 static void nvmet_bdev_execute_flush(struct nvmet_req *req) 339 { 340 struct bio *bio = &req->b.inline_bio; 341 342 if (!bdev_write_cache(req->ns->bdev)) { 343 nvmet_req_complete(req, NVME_SC_SUCCESS); 344 return; 345 } 346 347 if (!nvmet_check_transfer_len(req, 0)) 348 return; 349 350 bio_init(bio, req->ns->bdev, req->inline_bvec, 351 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH); 352 bio->bi_private = req; 353 bio->bi_end_io = nvmet_bio_done; 354 355 submit_bio(bio); 356 } 357 358 u16 nvmet_bdev_flush(struct nvmet_req *req) 359 { 360 if (!bdev_write_cache(req->ns->bdev)) 361 return 0; 362 363 if (blkdev_issue_flush(req->ns->bdev)) 364 return NVME_SC_INTERNAL | NVME_STATUS_DNR; 365 return 0; 366 } 367 368 static u16 nvmet_bdev_discard_range(struct nvmet_req *req, 369 struct nvme_dsm_range *range, struct bio **bio) 370 { 371 struct nvmet_ns *ns = req->ns; 372 int ret; 373 374 ret = __blkdev_issue_discard(ns->bdev, 375 nvmet_lba_to_sect(ns, range->slba), 376 le32_to_cpu(range->nlb) << (ns->blksize_shift - 9), 377 GFP_KERNEL, bio); 378 if (ret && ret != -EOPNOTSUPP) { 379 req->error_slba = le64_to_cpu(range->slba); 380 return errno_to_nvme_status(req, ret); 381 } 382 return NVME_SC_SUCCESS; 383 } 384 385 static void nvmet_bdev_execute_discard(struct nvmet_req *req) 386 { 387 struct nvme_dsm_range range; 388 struct bio *bio = NULL; 389 int i; 390 u16 status; 391 392 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) { 393 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range, 394 sizeof(range)); 395 if (status) 396 break; 397 398 status = nvmet_bdev_discard_range(req, &range, &bio); 399 if (status) 400 break; 401 } 402 403 if (bio) { 404 bio->bi_private = req; 405 bio->bi_end_io = nvmet_bio_done; 406 if (status) 407 bio_io_error(bio); 408 else 409 submit_bio(bio); 410 } else { 411 nvmet_req_complete(req, status); 412 } 413 } 414 415 static void nvmet_bdev_execute_dsm(struct nvmet_req *req) 416 { 417 if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req))) 418 return; 419 420 switch (le32_to_cpu(req->cmd->dsm.attributes)) { 421 case NVME_DSMGMT_AD: 422 nvmet_bdev_execute_discard(req); 423 return; 424 case NVME_DSMGMT_IDR: 425 case NVME_DSMGMT_IDW: 426 default: 427 /* Not supported yet */ 428 nvmet_req_complete(req, 0); 429 return; 430 } 431 } 432 433 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req) 434 { 435 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes; 436 struct bio *bio = NULL; 437 sector_t sector; 438 sector_t nr_sector; 439 int ret; 440 441 if (!nvmet_check_transfer_len(req, 0)) 442 return; 443 444 sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba); 445 nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) << 446 (req->ns->blksize_shift - 9)); 447 448 ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector, 449 GFP_KERNEL, &bio, 0); 450 if (bio) { 451 bio->bi_private = req; 452 bio->bi_end_io = nvmet_bio_done; 453 submit_bio(bio); 454 } else { 455 nvmet_req_complete(req, errno_to_nvme_status(req, ret)); 456 } 457 } 458 459 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req) 460 { 461 switch (req->cmd->common.opcode) { 462 case nvme_cmd_read: 463 case nvme_cmd_write: 464 req->execute = nvmet_bdev_execute_rw; 465 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns)) 466 req->metadata_len = nvmet_rw_metadata_len(req); 467 return 0; 468 case nvme_cmd_flush: 469 req->execute = nvmet_bdev_execute_flush; 470 return 0; 471 case nvme_cmd_dsm: 472 req->execute = nvmet_bdev_execute_dsm; 473 return 0; 474 case nvme_cmd_write_zeroes: 475 req->execute = nvmet_bdev_execute_write_zeroes; 476 return 0; 477 default: 478 return nvmet_report_invalid_opcode(req); 479 } 480 } 481