linux/drivers/nvdimm/blk.c
<<
>>
Prefs
   1/*
   2 * NVDIMM Block Window Driver
   3 * Copyright (c) 2014, Intel Corporation.
   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
  15#include <linux/blkdev.h>
  16#include <linux/fs.h>
  17#include <linux/genhd.h>
  18#include <linux/module.h>
  19#include <linux/moduleparam.h>
  20#include <linux/nd.h>
  21#include <linux/sizes.h>
  22#include "nd.h"
  23
  24static u32 nsblk_meta_size(struct nd_namespace_blk *nsblk)
  25{
  26        return nsblk->lbasize - ((nsblk->lbasize >= 4096) ? 4096 : 512);
  27}
  28
  29static u32 nsblk_internal_lbasize(struct nd_namespace_blk *nsblk)
  30{
  31        return roundup(nsblk->lbasize, INT_LBASIZE_ALIGNMENT);
  32}
  33
  34static u32 nsblk_sector_size(struct nd_namespace_blk *nsblk)
  35{
  36        return nsblk->lbasize - nsblk_meta_size(nsblk);
  37}
  38
  39static resource_size_t to_dev_offset(struct nd_namespace_blk *nsblk,
  40                                resource_size_t ns_offset, unsigned int len)
  41{
  42        int i;
  43
  44        for (i = 0; i < nsblk->num_resources; i++) {
  45                if (ns_offset < resource_size(nsblk->res[i])) {
  46                        if (ns_offset + len > resource_size(nsblk->res[i])) {
  47                                dev_WARN_ONCE(&nsblk->common.dev, 1,
  48                                        "illegal request\n");
  49                                return SIZE_MAX;
  50                        }
  51                        return nsblk->res[i]->start + ns_offset;
  52                }
  53                ns_offset -= resource_size(nsblk->res[i]);
  54        }
  55
  56        dev_WARN_ONCE(&nsblk->common.dev, 1, "request out of range\n");
  57        return SIZE_MAX;
  58}
  59
  60static struct nd_blk_region *to_ndbr(struct nd_namespace_blk *nsblk)
  61{
  62        struct nd_region *nd_region;
  63        struct device *parent;
  64
  65        parent = nsblk->common.dev.parent;
  66        nd_region = container_of(parent, struct nd_region, dev);
  67        return container_of(nd_region, struct nd_blk_region, nd_region);
  68}
  69
  70#ifdef CONFIG_BLK_DEV_INTEGRITY
  71static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
  72                struct bio_integrity_payload *bip, u64 lba, int rw)
  73{
  74        struct nd_blk_region *ndbr = to_ndbr(nsblk);
  75        unsigned int len = nsblk_meta_size(nsblk);
  76        resource_size_t dev_offset, ns_offset;
  77        u32 internal_lbasize, sector_size;
  78        struct bio_vec *bv;
  79        int err = 0, i;
  80
  81        internal_lbasize = nsblk_internal_lbasize(nsblk);
  82        sector_size = nsblk_sector_size(nsblk);
  83        ns_offset = lba * internal_lbasize + sector_size;
  84        dev_offset = to_dev_offset(nsblk, ns_offset, len);
  85        if (dev_offset == SIZE_MAX)
  86                return -EIO;
  87
  88        bip_for_each_vec(bv, bip, i) {
  89                unsigned int cur_len;
  90                void *iobuf;
  91
  92                if (!len)
  93                        break;
  94
  95                cur_len = min(len, bv->bv_len);
  96                iobuf = kmap_atomic(bv->bv_page);
  97                err = ndbr->do_io(ndbr, dev_offset, iobuf + bv->bv_offset,
  98                                cur_len, rw);
  99                kunmap_atomic(iobuf);
 100                if (err)
 101                        return err;
 102
 103                len -= cur_len;
 104                dev_offset += cur_len;
 105        }
 106        BUG_ON(len);
 107
 108        return err;
 109}
 110
 111#else /* CONFIG_BLK_DEV_INTEGRITY */
 112static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
 113                struct bio_integrity_payload *bip, u64 lba, int rw)
 114{
 115        return 0;
 116}
 117#endif
 118
 119static int nsblk_do_bvec(struct nd_namespace_blk *nsblk,
 120                struct bio_integrity_payload *bip, struct page *page,
 121                unsigned int len, unsigned int off, int rw, sector_t sector)
 122{
 123        struct nd_blk_region *ndbr = to_ndbr(nsblk);
 124        resource_size_t dev_offset, ns_offset;
 125        u32 internal_lbasize, sector_size;
 126        int err = 0;
 127        void *iobuf;
 128        u64 lba;
 129
 130        internal_lbasize = nsblk_internal_lbasize(nsblk);
 131        sector_size = nsblk_sector_size(nsblk);
 132        while (len) {
 133                unsigned int cur_len;
 134
 135                /*
 136                 * If we don't have an integrity payload, we don't have to
 137                 * split the bvec into sectors, as this would cause unnecessary
 138                 * Block Window setup/move steps. the do_io routine is capable
 139                 * of handling len <= PAGE_SIZE.
 140                 */
 141                cur_len = bip ? min(len, sector_size) : len;
 142
 143                lba = div_u64(sector << SECTOR_SHIFT, sector_size);
 144                ns_offset = lba * internal_lbasize;
 145                dev_offset = to_dev_offset(nsblk, ns_offset, cur_len);
 146                if (dev_offset == SIZE_MAX)
 147                        return -EIO;
 148
 149                iobuf = kmap_atomic(page);
 150                err = ndbr->do_io(ndbr, dev_offset, iobuf + off, cur_len, rw);
 151                kunmap_atomic(iobuf);
 152                if (err)
 153                        return err;
 154
 155                if (bip) {
 156                        err = nd_blk_rw_integrity(nsblk, bip, lba, rw);
 157                        if (err)
 158                                return err;
 159                }
 160                len -= cur_len;
 161                off += cur_len;
 162                sector += sector_size >> SECTOR_SHIFT;
 163        }
 164
 165        return err;
 166}
 167
 168static void nd_blk_make_request(struct request_queue *q, struct bio *bio)
 169{
 170        struct bio_integrity_payload *bip;
 171        struct nd_namespace_blk *nsblk;
 172        struct bio_vec *bvec;
 173        unsigned long start;
 174        sector_t sector = bio->bi_sector;
 175        int err = 0, rw, i;
 176        bool do_acct;
 177
 178        /*
 179         * bio_integrity_enabled also checks if the bio already has an
 180         * integrity payload attached. If it does, we *don't* do a
 181         * bio_integrity_prep here - the payload has been generated by
 182         * another kernel subsystem, and we just pass it through.
 183         */
 184        if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
 185                err = -EIO;
 186                goto out;
 187        }
 188
 189        bip = bio->bi_integrity;
 190        nsblk = q->queuedata;
 191        rw = bio_data_dir(bio);
 192        do_acct = nd_iostat_start(bio, &start);
 193        bio_for_each_segment(bvec, bio, i) {
 194                unsigned int len = bvec->bv_len;
 195
 196                BUG_ON(len > PAGE_SIZE);
 197                err = nsblk_do_bvec(nsblk, bip, bvec->bv_page, len,
 198                                bvec->bv_offset, rw, sector);
 199                if (err) {
 200                        dev_dbg(&nsblk->common.dev,
 201                                        "io error in %s sector %lld, len %d,\n",
 202                                        (rw == READ) ? "READ" : "WRITE",
 203                                        (unsigned long long) sector, len);
 204                        break;
 205                }
 206                sector += len >> SECTOR_SHIFT;
 207        }
 208        if (do_acct)
 209                nd_iostat_end(bio, start);
 210
 211 out:
 212        bio_endio(bio, err);
 213}
 214
 215static int nsblk_rw_bytes(struct nd_namespace_common *ndns,
 216                resource_size_t offset, void *iobuf, size_t n, int rw)
 217{
 218        struct nd_namespace_blk *nsblk = to_nd_namespace_blk(&ndns->dev);
 219        struct nd_blk_region *ndbr = to_ndbr(nsblk);
 220        resource_size_t dev_offset;
 221
 222        dev_offset = to_dev_offset(nsblk, offset, n);
 223
 224        if (unlikely(offset + n > nsblk->size)) {
 225                dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
 226                return -EFAULT;
 227        }
 228
 229        if (dev_offset == SIZE_MAX)
 230                return -EIO;
 231
 232        return ndbr->do_io(ndbr, dev_offset, iobuf, n, rw);
 233}
 234
 235static const struct block_device_operations nd_blk_fops = {
 236        .owner = THIS_MODULE,
 237        .revalidate_disk = nvdimm_revalidate_disk,
 238};
 239
 240static void nd_blk_release_queue(void *q)
 241{
 242        blk_cleanup_queue(q);
 243}
 244
 245static void nd_blk_release_disk(void *disk)
 246{
 247        del_gendisk(disk);
 248        put_disk(disk);
 249}
 250
 251static int nsblk_attach_disk(struct nd_namespace_blk *nsblk)
 252{
 253        struct device *dev = &nsblk->common.dev;
 254        resource_size_t available_disk_size;
 255        struct request_queue *q;
 256        struct gendisk *disk;
 257        u64 internal_nlba;
 258
 259        internal_nlba = div_u64(nsblk->size, nsblk_internal_lbasize(nsblk));
 260        available_disk_size = internal_nlba * nsblk_sector_size(nsblk);
 261
 262        q = blk_alloc_queue(GFP_KERNEL);
 263        if (!q)
 264                return -ENOMEM;
 265        if (devm_add_action(dev, nd_blk_release_queue, q)) {
 266                blk_cleanup_queue(q);
 267                return -ENOMEM;
 268        }
 269
 270        blk_queue_make_request(q, nd_blk_make_request);
 271        blk_queue_max_hw_sectors(q, UINT_MAX);
 272        blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 273        blk_queue_logical_block_size(q, nsblk_sector_size(nsblk));
 274        queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
 275        q->queuedata = nsblk;
 276
 277        disk = alloc_disk(0);
 278        if (!disk)
 279                return -ENOMEM;
 280        if (devm_add_action(dev, nd_blk_release_disk, disk)) {
 281                put_disk(disk);
 282                return -ENOMEM;
 283        }
 284
 285        disk->driverfs_dev      = dev;
 286        disk->first_minor       = 0;
 287        disk->fops              = &nd_blk_fops;
 288        disk->queue             = q;
 289        disk->flags             = GENHD_FL_EXT_DEVT;
 290        nvdimm_namespace_disk_name(&nsblk->common, disk->disk_name);
 291        set_capacity(disk, 0);
 292        add_disk(disk);
 293
 294        if (nsblk_meta_size(nsblk)) {
 295                int rc = nd_integrity_init(disk, nsblk_meta_size(nsblk));
 296
 297                if (rc)
 298                        return rc;
 299        }
 300
 301        set_capacity(disk, available_disk_size >> SECTOR_SHIFT);
 302        revalidate_disk(disk);
 303        return 0;
 304}
 305
 306static int nd_blk_probe(struct device *dev)
 307{
 308        struct nd_namespace_common *ndns;
 309        struct nd_namespace_blk *nsblk;
 310
 311        ndns = nvdimm_namespace_common_probe(dev);
 312        if (IS_ERR(ndns))
 313                return PTR_ERR(ndns);
 314
 315        nsblk = to_nd_namespace_blk(&ndns->dev);
 316        nsblk->size = nvdimm_namespace_capacity(ndns);
 317        dev_set_drvdata(dev, nsblk);
 318
 319        ndns->rw_bytes = nsblk_rw_bytes;
 320        if (is_nd_btt(dev))
 321                return nvdimm_namespace_attach_btt(ndns);
 322        else if (nd_btt_probe(dev, ndns) == 0) {
 323                /* we'll come back as btt-blk */
 324                return -ENXIO;
 325        } else
 326                return nsblk_attach_disk(nsblk);
 327}
 328
 329static int nd_blk_remove(struct device *dev)
 330{
 331        if (is_nd_btt(dev))
 332                nvdimm_namespace_detach_btt(to_nd_btt(dev));
 333        return 0;
 334}
 335
 336static struct nd_device_driver nd_blk_driver = {
 337        .probe = nd_blk_probe,
 338        .remove = nd_blk_remove,
 339        .drv = {
 340                .name = "nd_blk",
 341        },
 342        .type = ND_DRIVER_NAMESPACE_BLK,
 343};
 344
 345static int __init nd_blk_init(void)
 346{
 347        return nd_driver_register(&nd_blk_driver);
 348}
 349
 350static void __exit nd_blk_exit(void)
 351{
 352        driver_unregister(&nd_blk_driver.drv);
 353}
 354
 355MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
 356MODULE_LICENSE("GPL v2");
 357MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_BLK);
 358module_init(nd_blk_init);
 359module_exit(nd_blk_exit);
 360