drivers/nvdimm/blk.c

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