drivers/nvme/target/io-cmd-bdev.c

   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/module.h>
   9 #include "nvmet.h"
  10
  11 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)
  12 {
  13         const struct queue_limits *ql = &bdev_get_queue(bdev)->limits;
  14         /* Number of logical blocks per physical block. */
  15         const u32 lpp = ql->physical_block_size / ql->logical_block_size;
  16         /* Logical blocks per physical block, 0's based. */
  17         const __le16 lpp0b = to0based(lpp);
  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 = lpp0b;
  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(ql->discard_granularity / ql->logical_block_size);
  44         /* NPDG = Namespace Preferred Deallocate Alignment */
  45         id->npda = id->npdg;
  46         /* NOWS = Namespace Optimal Write Size */
  47         id->nows = to0based(ql->io_opt / ql->logical_block_size);
  48 }
  49
  50 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns)
  51 {
  52         struct blk_integrity *bi = bdev_get_integrity(ns->bdev);
  53
  54         if (bi) {
  55                 ns->metadata_size = bi->tuple_size;
  56                 if (bi->profile == &t10_pi_type1_crc)
  57                         ns->pi_type = NVME_NS_DPS_PI_TYPE1;
  58                 else if (bi->profile == &t10_pi_type3_crc)
  59                         ns->pi_type = NVME_NS_DPS_PI_TYPE3;
  60                 else
  61                         /* Unsupported metadata type */
  62                         ns->metadata_size = 0;
  63         }
  64 }
  65
  66 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
  67 {
  68         int ret;
  69
  70         ns->bdev = blkdev_get_by_path(ns->device_path,
  71                         FMODE_READ | FMODE_WRITE, NULL);
  72         if (IS_ERR(ns->bdev)) {
  73                 ret = PTR_ERR(ns->bdev);
  74                 if (ret != -ENOTBLK) {
  75                         pr_err("failed to open block device %s: (%ld)\n",
  76                                         ns->device_path, PTR_ERR(ns->bdev));
  77                 }
  78                 ns->bdev = NULL;
  79                 return ret;
  80         }
  81         ns->size = i_size_read(ns->bdev->bd_inode);
  82         ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
  83
  84         ns->pi_type = 0;
  85         ns->metadata_size = 0;
  86         if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY_T10))
  87                 nvmet_bdev_ns_enable_integrity(ns);
  88
  89         return 0;
  90 }
  91
  92 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
  93 {
  94         if (ns->bdev) {
  95                 blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
  96                 ns->bdev = NULL;
  97         }
  98 }
  99
 100 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
 101 {
 102         ns->size = i_size_read(ns->bdev->bd_inode);
 103 }
 104
 105 static u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
 106 {
 107         u16 status = NVME_SC_SUCCESS;
 108
 109         if (likely(blk_sts == BLK_STS_OK))
 110                 return status;
 111         /*
 112          * Right now there exists M : 1 mapping between block layer error
 113          * to the NVMe status code (see nvme_error_status()). For consistency,
 114          * when we reverse map we use most appropriate NVMe Status code from
 115          * the group of the NVMe staus codes used in the nvme_error_status().
 116          */
 117         switch (blk_sts) {
 118         case BLK_STS_NOSPC:
 119                 status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
 120                 req->error_loc = offsetof(struct nvme_rw_command, length);
 121                 break;
 122         case BLK_STS_TARGET:
 123                 status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
 124                 req->error_loc = offsetof(struct nvme_rw_command, slba);
 125                 break;
 126         case BLK_STS_NOTSUPP:
 127                 req->error_loc = offsetof(struct nvme_common_command, opcode);
 128                 switch (req->cmd->common.opcode) {
 129                 case nvme_cmd_dsm:
 130                 case nvme_cmd_write_zeroes:
 131                         status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
 132                         break;
 133                 default:
 134                         status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
 135                 }
 136                 break;
 137         case BLK_STS_MEDIUM:
 138                 status = NVME_SC_ACCESS_DENIED;
 139                 req->error_loc = offsetof(struct nvme_rw_command, nsid);
 140                 break;
 141         case BLK_STS_IOERR:
 142         default:
 143                 status = NVME_SC_INTERNAL | NVME_SC_DNR;
 144                 req->error_loc = offsetof(struct nvme_common_command, opcode);
 145         }
 146
 147         switch (req->cmd->common.opcode) {
 148         case nvme_cmd_read:
 149         case nvme_cmd_write:
 150                 req->error_slba = le64_to_cpu(req->cmd->rw.slba);
 151                 break;
 152         case nvme_cmd_write_zeroes:
 153                 req->error_slba =
 154                         le64_to_cpu(req->cmd->write_zeroes.slba);
 155                 break;
 156         default:
 157                 req->error_slba = 0;
 158         }
 159         return status;
 160 }
 161
 162 static void nvmet_bio_done(struct bio *bio)
 163 {
 164         struct nvmet_req *req = bio->bi_private;
 165
 166         nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
 167         if (bio != &req->b.inline_bio)
 168                 bio_put(bio);
 169 }
 170
 171 #ifdef CONFIG_BLK_DEV_INTEGRITY
 172 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
 173                                 struct sg_mapping_iter *miter)
 174 {
 175         struct blk_integrity *bi;
 176         struct bio_integrity_payload *bip;
 177         int rc;
 178         size_t resid, len;
 179
 180         bi = bdev_get_integrity(req->ns->bdev);
 181         if (unlikely(!bi)) {
 182                 pr_err("Unable to locate bio_integrity\n");
 183                 return -ENODEV;
 184         }
 185
 186         bip = bio_integrity_alloc(bio, GFP_NOIO,
 187                                         bio_max_segs(req->metadata_sg_cnt));
 188         if (IS_ERR(bip)) {
 189                 pr_err("Unable to allocate bio_integrity_payload\n");
 190                 return PTR_ERR(bip);
 191         }
 192
 193         bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
 194         /* virtual start sector must be in integrity interval units */
 195         bip_set_seed(bip, bio->bi_iter.bi_sector >>
 196                      (bi->interval_exp - SECTOR_SHIFT));
 197
 198         resid = bip->bip_iter.bi_size;
 199         while (resid > 0 && sg_miter_next(miter)) {
 200                 len = min_t(size_t, miter->length, resid);
 201                 rc = bio_integrity_add_page(bio, miter->page, len,
 202                                             offset_in_page(miter->addr));
 203                 if (unlikely(rc != len)) {
 204                         pr_err("bio_integrity_add_page() failed; %d\n", rc);
 205                         sg_miter_stop(miter);
 206                         return -ENOMEM;
 207                 }
 208
 209                 resid -= len;
 210                 if (len < miter->length)
 211                         miter->consumed -= miter->length - len;
 212         }
 213         sg_miter_stop(miter);
 214
 215         return 0;
 216 }
 217 #else
 218 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
 219                                 struct sg_mapping_iter *miter)
 220 {
 221         return -EINVAL;
 222 }
 223 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 224
 225 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
 226 {
 227         unsigned int sg_cnt = req->sg_cnt;
 228         struct bio *bio;
 229         struct scatterlist *sg;
 230         struct blk_plug plug;
 231         sector_t sector;
 232         int op, i, rc;
 233         struct sg_mapping_iter prot_miter;
 234         unsigned int iter_flags;
 235         unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
 236
 237         if (!nvmet_check_transfer_len(req, total_len))
 238                 return;
 239
 240         if (!req->sg_cnt) {
 241                 nvmet_req_complete(req, 0);
 242                 return;
 243         }
 244
 245         if (req->cmd->rw.opcode == nvme_cmd_write) {
 246                 op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
 247                 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
 248                         op |= REQ_FUA;
 249                 iter_flags = SG_MITER_TO_SG;
 250         } else {
 251                 op = REQ_OP_READ;
 252                 iter_flags = SG_MITER_FROM_SG;
 253         }
 254
 255         if (is_pci_p2pdma_page(sg_page(req->sg)))
 256                 op |= REQ_NOMERGE;
 257
 258         sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
 259
 260         if (nvmet_use_inline_bvec(req)) {
 261                 bio = &req->b.inline_bio;
 262                 bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
 263         } else {
 264                 bio = bio_alloc(GFP_KERNEL, bio_max_segs(sg_cnt));
 265         }
 266         bio_set_dev(bio, req->ns->bdev);
 267         bio->bi_iter.bi_sector = sector;
 268         bio->bi_private = req;
 269         bio->bi_end_io = nvmet_bio_done;
 270         bio->bi_opf = op;
 271
 272         blk_start_plug(&plug);
 273         if (req->metadata_len)
 274                 sg_miter_start(&prot_miter, req->metadata_sg,
 275                                req->metadata_sg_cnt, iter_flags);
 276
 277         for_each_sg(req->sg, sg, req->sg_cnt, i) {
 278                 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
 279                                 != sg->length) {
 280                         struct bio *prev = bio;
 281
 282                         if (req->metadata_len) {
 283                                 rc = nvmet_bdev_alloc_bip(req, bio,
 284                                                           &prot_miter);
 285                                 if (unlikely(rc)) {
 286                                         bio_io_error(bio);
 287                                         return;
 288                                 }
 289                         }
 290
 291                         bio = bio_alloc(GFP_KERNEL, bio_max_segs(sg_cnt));
 292                         bio_set_dev(bio, req->ns->bdev);
 293                         bio->bi_iter.bi_sector = sector;
 294                         bio->bi_opf = op;
 295
 296                         bio_chain(bio, prev);
 297                         submit_bio(prev);
 298                 }
 299
 300                 sector += sg->length >> 9;
 301                 sg_cnt--;
 302         }
 303
 304         if (req->metadata_len) {
 305                 rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
 306                 if (unlikely(rc)) {
 307                         bio_io_error(bio);
 308                         return;
 309                 }
 310         }
 311
 312         submit_bio(bio);
 313         blk_finish_plug(&plug);
 314 }
 315
 316 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
 317 {
 318         struct bio *bio = &req->b.inline_bio;
 319
 320         if (!nvmet_check_transfer_len(req, 0))
 321                 return;
 322
 323         bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
 324         bio_set_dev(bio, req->ns->bdev);
 325         bio->bi_private = req;
 326         bio->bi_end_io = nvmet_bio_done;
 327         bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
 328
 329         submit_bio(bio);
 330 }
 331
 332 u16 nvmet_bdev_flush(struct nvmet_req *req)
 333 {
 334         if (blkdev_issue_flush(req->ns->bdev))
 335                 return NVME_SC_INTERNAL | NVME_SC_DNR;
 336         return 0;
 337 }
 338
 339 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
 340                 struct nvme_dsm_range *range, struct bio **bio)
 341 {
 342         struct nvmet_ns *ns = req->ns;
 343         int ret;
 344
 345         ret = __blkdev_issue_discard(ns->bdev,
 346                         nvmet_lba_to_sect(ns, range->slba),
 347                         le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
 348                         GFP_KERNEL, 0, bio);
 349         if (ret && ret != -EOPNOTSUPP) {
 350                 req->error_slba = le64_to_cpu(range->slba);
 351                 return errno_to_nvme_status(req, ret);
 352         }
 353         return NVME_SC_SUCCESS;
 354 }
 355
 356 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
 357 {
 358         struct nvme_dsm_range range;
 359         struct bio *bio = NULL;
 360         int i;
 361         u16 status;
 362
 363         for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
 364                 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
 365                                 sizeof(range));
 366                 if (status)
 367                         break;
 368
 369                 status = nvmet_bdev_discard_range(req, &range, &bio);
 370                 if (status)
 371                         break;
 372         }
 373
 374         if (bio) {
 375                 bio->bi_private = req;
 376                 bio->bi_end_io = nvmet_bio_done;
 377                 if (status)
 378                         bio_io_error(bio);
 379                 else
 380                         submit_bio(bio);
 381         } else {
 382                 nvmet_req_complete(req, status);
 383         }
 384 }
 385
 386 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
 387 {
 388         if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
 389                 return;
 390
 391         switch (le32_to_cpu(req->cmd->dsm.attributes)) {
 392         case NVME_DSMGMT_AD:
 393                 nvmet_bdev_execute_discard(req);
 394                 return;
 395         case NVME_DSMGMT_IDR:
 396         case NVME_DSMGMT_IDW:
 397         default:
 398                 /* Not supported yet */
 399                 nvmet_req_complete(req, 0);
 400                 return;
 401         }
 402 }
 403
 404 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
 405 {
 406         struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
 407         struct bio *bio = NULL;
 408         sector_t sector;
 409         sector_t nr_sector;
 410         int ret;
 411
 412         if (!nvmet_check_transfer_len(req, 0))
 413                 return;
 414
 415         sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
 416         nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
 417                 (req->ns->blksize_shift - 9));
 418
 419         ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
 420                         GFP_KERNEL, &bio, 0);
 421         if (bio) {
 422                 bio->bi_private = req;
 423                 bio->bi_end_io = nvmet_bio_done;
 424                 submit_bio(bio);
 425         } else {
 426                 nvmet_req_complete(req, errno_to_nvme_status(req, ret));
 427         }
 428 }
 429
 430 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
 431 {
 432         switch (req->cmd->common.opcode) {
 433         case nvme_cmd_read:
 434         case nvme_cmd_write:
 435                 req->execute = nvmet_bdev_execute_rw;
 436                 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
 437                         req->metadata_len = nvmet_rw_metadata_len(req);
 438                 return 0;
 439         case nvme_cmd_flush:
 440                 req->execute = nvmet_bdev_execute_flush;
 441                 return 0;
 442         case nvme_cmd_dsm:
 443                 req->execute = nvmet_bdev_execute_dsm;
 444                 return 0;
 445         case nvme_cmd_write_zeroes:
 446                 req->execute = nvmet_bdev_execute_write_zeroes;
 447                 return 0;
 448         default:
 449                 return nvmet_report_invalid_opcode(req);
 450         }
 451 }