fs/nfs/blocklayout/dev.c

   1 /*
   2  * Copyright (c) 2014-2016 Christoph Hellwig.
   3  */
   4 #include <linux/sunrpc/svc.h>
   5 #include <linux/blkdev.h>
   6 #include <linux/nfs4.h>
   7 #include <linux/nfs_fs.h>
   8 #include <linux/nfs_xdr.h>
   9 #include <linux/pr.h>
  10
  11 #include "blocklayout.h"
  12
  13 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  14
  15 static void
  16 bl_free_device(struct pnfs_block_dev *dev)
  17 {
  18         if (dev->nr_children) {
  19                 int i;
  20
  21                 for (i = 0; i < dev->nr_children; i++)
  22                         bl_free_device(&dev->children[i]);
  23                 kfree(dev->children);
  24         } else {
  25                 if (dev->pr_registered) {
  26                         const struct pr_ops *ops =
  27                                 dev->bdev->bd_disk->fops->pr_ops;
  28                         int error;
  29
  30                         error = ops->pr_register(dev->bdev, dev->pr_key, 0,
  31                                 false);
  32                         if (error)
  33                                 pr_err("failed to unregister PR key.\n");
  34                 }
  35
  36                 if (dev->bdev)
  37                         blkdev_put(dev->bdev, FMODE_READ | FMODE_WRITE);
  38         }
  39 }
  40
  41 void
  42 bl_free_deviceid_node(struct nfs4_deviceid_node *d)
  43 {
  44         struct pnfs_block_dev *dev =
  45                 container_of(d, struct pnfs_block_dev, node);
  46
  47         bl_free_device(dev);
  48         kfree_rcu(dev, node.rcu);
  49 }
  50
  51 static int
  52 nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b)
  53 {
  54         __be32 *p;
  55         int i;
  56
  57         p = xdr_inline_decode(xdr, 4);
  58         if (!p)
  59                 return -EIO;
  60         b->type = be32_to_cpup(p++);
  61
  62         switch (b->type) {
  63         case PNFS_BLOCK_VOLUME_SIMPLE:
  64                 p = xdr_inline_decode(xdr, 4);
  65                 if (!p)
  66                         return -EIO;
  67                 b->simple.nr_sigs = be32_to_cpup(p++);
  68                 if (!b->simple.nr_sigs) {
  69                         dprintk("no signature\n");
  70                         return -EIO;
  71                 }
  72
  73                 b->simple.len = 4 + 4;
  74                 for (i = 0; i < b->simple.nr_sigs; i++) {
  75                         p = xdr_inline_decode(xdr, 8 + 4);
  76                         if (!p)
  77                                 return -EIO;
  78                         p = xdr_decode_hyper(p, &b->simple.sigs[i].offset);
  79                         b->simple.sigs[i].sig_len = be32_to_cpup(p++);
  80                         if (b->simple.sigs[i].sig_len > PNFS_BLOCK_UUID_LEN) {
  81                                 pr_info("signature too long: %d\n",
  82                                         b->simple.sigs[i].sig_len);
  83                                 return -EIO;
  84                         }
  85
  86                         p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
  87                         if (!p)
  88                                 return -EIO;
  89                         memcpy(&b->simple.sigs[i].sig, p,
  90                                 b->simple.sigs[i].sig_len);
  91
  92                         b->simple.len += 8 + 4 + b->simple.sigs[i].sig_len;
  93                 }
  94                 break;
  95         case PNFS_BLOCK_VOLUME_SLICE:
  96                 p = xdr_inline_decode(xdr, 8 + 8 + 4);
  97                 if (!p)
  98                         return -EIO;
  99                 p = xdr_decode_hyper(p, &b->slice.start);
 100                 p = xdr_decode_hyper(p, &b->slice.len);
 101                 b->slice.volume = be32_to_cpup(p++);
 102                 break;
 103         case PNFS_BLOCK_VOLUME_CONCAT:
 104                 p = xdr_inline_decode(xdr, 4);
 105                 if (!p)
 106                         return -EIO;
 107                 b->concat.volumes_count = be32_to_cpup(p++);
 108
 109                 p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
 110                 if (!p)
 111                         return -EIO;
 112                 for (i = 0; i < b->concat.volumes_count; i++)
 113                         b->concat.volumes[i] = be32_to_cpup(p++);
 114                 break;
 115         case PNFS_BLOCK_VOLUME_STRIPE:
 116                 p = xdr_inline_decode(xdr, 8 + 4);
 117                 if (!p)
 118                         return -EIO;
 119                 p = xdr_decode_hyper(p, &b->stripe.chunk_size);
 120                 b->stripe.volumes_count = be32_to_cpup(p++);
 121
 122                 p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
 123                 if (!p)
 124                         return -EIO;
 125                 for (i = 0; i < b->stripe.volumes_count; i++)
 126                         b->stripe.volumes[i] = be32_to_cpup(p++);
 127                 break;
 128         case PNFS_BLOCK_VOLUME_SCSI:
 129                 p = xdr_inline_decode(xdr, 4 + 4 + 4);
 130                 if (!p)
 131                         return -EIO;
 132                 b->scsi.code_set = be32_to_cpup(p++);
 133                 b->scsi.designator_type = be32_to_cpup(p++);
 134                 b->scsi.designator_len = be32_to_cpup(p++);
 135                 p = xdr_inline_decode(xdr, b->scsi.designator_len);
 136                 if (!p)
 137                         return -EIO;
 138                 if (b->scsi.designator_len > 256)
 139                         return -EIO;
 140                 memcpy(&b->scsi.designator, p, b->scsi.designator_len);
 141                 p = xdr_inline_decode(xdr, 8);
 142                 if (!p)
 143                         return -EIO;
 144                 p = xdr_decode_hyper(p, &b->scsi.pr_key);
 145                 break;
 146         default:
 147                 dprintk("unknown volume type!\n");
 148                 return -EIO;
 149         }
 150
 151         return 0;
 152 }
 153
 154 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
 155                 struct pnfs_block_dev_map *map)
 156 {
 157         map->start = dev->start;
 158         map->len = dev->len;
 159         map->disk_offset = dev->disk_offset;
 160         map->bdev = dev->bdev;
 161         return true;
 162 }
 163
 164 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
 165                 struct pnfs_block_dev_map *map)
 166 {
 167         int i;
 168
 169         for (i = 0; i < dev->nr_children; i++) {
 170                 struct pnfs_block_dev *child = &dev->children[i];
 171
 172                 if (child->start > offset ||
 173                     child->start + child->len <= offset)
 174                         continue;
 175
 176                 child->map(child, offset - child->start, map);
 177                 return true;
 178         }
 179
 180         dprintk("%s: ran off loop!\n", __func__);
 181         return false;
 182 }
 183
 184 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
 185                 struct pnfs_block_dev_map *map)
 186 {
 187         struct pnfs_block_dev *child;
 188         u64 chunk;
 189         u32 chunk_idx;
 190         u64 disk_offset;
 191
 192         chunk = div_u64(offset, dev->chunk_size);
 193         div_u64_rem(chunk, dev->nr_children, &chunk_idx);
 194
 195         if (chunk_idx > dev->nr_children) {
 196                 dprintk("%s: invalid chunk idx %d (%lld/%lld)\n",
 197                         __func__, chunk_idx, offset, dev->chunk_size);
 198                 /* error, should not happen */
 199                 return false;
 200         }
 201
 202         /* truncate offset to the beginning of the stripe */
 203         offset = chunk * dev->chunk_size;
 204
 205         /* disk offset of the stripe */
 206         disk_offset = div_u64(offset, dev->nr_children);
 207
 208         child = &dev->children[chunk_idx];
 209         child->map(child, disk_offset, map);
 210
 211         map->start += offset;
 212         map->disk_offset += disk_offset;
 213         map->len = dev->chunk_size;
 214         return true;
 215 }
 216
 217 static int
 218 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
 219                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
 220
 221
 222 static int
 223 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
 224                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
 225 {
 226         struct pnfs_block_volume *v = &volumes[idx];
 227         dev_t dev;
 228
 229         dev = bl_resolve_deviceid(server, v, gfp_mask);
 230         if (!dev)
 231                 return -EIO;
 232
 233         d->bdev = blkdev_get_by_dev(dev, FMODE_READ | FMODE_WRITE, NULL);
 234         if (IS_ERR(d->bdev)) {
 235                 printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
 236                         MAJOR(dev), MINOR(dev), PTR_ERR(d->bdev));
 237                 return PTR_ERR(d->bdev);
 238         }
 239
 240
 241         d->len = i_size_read(d->bdev->bd_inode);
 242         d->map = bl_map_simple;
 243
 244         printk(KERN_INFO "pNFS: using block device %s\n",
 245                 d->bdev->bd_disk->disk_name);
 246         return 0;
 247 }
 248
 249 static bool
 250 bl_validate_designator(struct pnfs_block_volume *v)
 251 {
 252         switch (v->scsi.designator_type) {
 253         case PS_DESIGNATOR_EUI64:
 254                 if (v->scsi.code_set != PS_CODE_SET_BINARY)
 255                         return false;
 256
 257                 if (v->scsi.designator_len != 8 &&
 258                     v->scsi.designator_len != 10 &&
 259                     v->scsi.designator_len != 16)
 260                         return false;
 261
 262                 return true;
 263         case PS_DESIGNATOR_NAA:
 264                 if (v->scsi.code_set != PS_CODE_SET_BINARY)
 265                         return false;
 266
 267                 if (v->scsi.designator_len != 8 &&
 268                     v->scsi.designator_len != 16)
 269                         return false;
 270
 271                 return true;
 272         case PS_DESIGNATOR_T10:
 273         case PS_DESIGNATOR_NAME:
 274                 pr_err("pNFS: unsupported designator "
 275                         "(code set %d, type %d, len %d.\n",
 276                         v->scsi.code_set,
 277                         v->scsi.designator_type,
 278                         v->scsi.designator_len);
 279                 return false;
 280         default:
 281                 pr_err("pNFS: invalid designator "
 282                         "(code set %d, type %d, len %d.\n",
 283                         v->scsi.code_set,
 284                         v->scsi.designator_type,
 285                         v->scsi.designator_len);
 286                 return false;
 287         }
 288 }
 289
 290 /*
 291  * Try to open the udev path for the WWN.  At least on Debian the udev
 292  * by-id path will always point to the dm-multipath device if one exists.
 293  */
 294 static struct block_device *
 295 bl_open_udev_path(struct pnfs_block_volume *v)
 296 {
 297         struct block_device *bdev;
 298         const char *devname;
 299
 300         devname = kasprintf(GFP_KERNEL, "/dev/disk/by-id/wwn-0x%*phN",
 301                                 v->scsi.designator_len, v->scsi.designator);
 302         if (!devname)
 303                 return ERR_PTR(-ENOMEM);
 304
 305         bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
 306         if (IS_ERR(bdev)) {
 307                 pr_warn("pNFS: failed to open device %s (%ld)\n",
 308                         devname, PTR_ERR(bdev));
 309         }
 310
 311         kfree(devname);
 312         return bdev;
 313 }
 314
 315 static int
 316 bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d,
 317                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
 318 {
 319         struct pnfs_block_volume *v = &volumes[idx];
 320         const struct pr_ops *ops;
 321         int error;
 322
 323         if (!bl_validate_designator(v))
 324                 return -EINVAL;
 325
 326         d->bdev = bl_open_udev_path(v);
 327         if (IS_ERR(d->bdev))
 328                 return PTR_ERR(d->bdev);
 329
 330         d->len = i_size_read(d->bdev->bd_inode);
 331         d->map = bl_map_simple;
 332         d->pr_key = v->scsi.pr_key;
 333
 334         pr_info("pNFS: using block device %s (reservation key 0x%llx)\n",
 335                 d->bdev->bd_disk->disk_name, d->pr_key);
 336
 337         ops = d->bdev->bd_disk->fops->pr_ops;
 338         if (!ops) {
 339                 pr_err("pNFS: block device %s does not support reservations.",
 340                                 d->bdev->bd_disk->disk_name);
 341                 error = -EINVAL;
 342                 goto out_blkdev_put;
 343         }
 344
 345         error = ops->pr_register(d->bdev, 0, d->pr_key, true);
 346         if (error) {
 347                 pr_err("pNFS: failed to register key for block device %s.",
 348                                 d->bdev->bd_disk->disk_name);
 349                 goto out_blkdev_put;
 350         }
 351
 352         d->pr_registered = true;
 353         return 0;
 354
 355 out_blkdev_put:
 356         blkdev_put(d->bdev, FMODE_READ | FMODE_WRITE);
 357         return error;
 358 }
 359
 360 static int
 361 bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d,
 362                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
 363 {
 364         struct pnfs_block_volume *v = &volumes[idx];
 365         int ret;
 366
 367         ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
 368         if (ret)
 369                 return ret;
 370
 371         d->disk_offset = v->slice.start;
 372         d->len = v->slice.len;
 373         return 0;
 374 }
 375
 376 static int
 377 bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d,
 378                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
 379 {
 380         struct pnfs_block_volume *v = &volumes[idx];
 381         u64 len = 0;
 382         int ret, i;
 383
 384         d->children = kcalloc(v->concat.volumes_count,
 385                         sizeof(struct pnfs_block_dev), GFP_KERNEL);
 386         if (!d->children)
 387                 return -ENOMEM;
 388
 389         for (i = 0; i < v->concat.volumes_count; i++) {
 390                 ret = bl_parse_deviceid(server, &d->children[i],
 391                                 volumes, v->concat.volumes[i], gfp_mask);
 392                 if (ret)
 393                         return ret;
 394
 395                 d->nr_children++;
 396                 d->children[i].start += len;
 397                 len += d->children[i].len;
 398         }
 399
 400         d->len = len;
 401         d->map = bl_map_concat;
 402         return 0;
 403 }
 404
 405 static int
 406 bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d,
 407                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
 408 {
 409         struct pnfs_block_volume *v = &volumes[idx];
 410         u64 len = 0;
 411         int ret, i;
 412
 413         d->children = kcalloc(v->stripe.volumes_count,
 414                         sizeof(struct pnfs_block_dev), GFP_KERNEL);
 415         if (!d->children)
 416                 return -ENOMEM;
 417
 418         for (i = 0; i < v->stripe.volumes_count; i++) {
 419                 ret = bl_parse_deviceid(server, &d->children[i],
 420                                 volumes, v->stripe.volumes[i], gfp_mask);
 421                 if (ret)
 422                         return ret;
 423
 424                 d->nr_children++;
 425                 len += d->children[i].len;
 426         }
 427
 428         d->len = len;
 429         d->chunk_size = v->stripe.chunk_size;
 430         d->map = bl_map_stripe;
 431         return 0;
 432 }
 433
 434 static int
 435 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
 436                 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
 437 {
 438         switch (volumes[idx].type) {
 439         case PNFS_BLOCK_VOLUME_SIMPLE:
 440                 return bl_parse_simple(server, d, volumes, idx, gfp_mask);
 441         case PNFS_BLOCK_VOLUME_SLICE:
 442                 return bl_parse_slice(server, d, volumes, idx, gfp_mask);
 443         case PNFS_BLOCK_VOLUME_CONCAT:
 444                 return bl_parse_concat(server, d, volumes, idx, gfp_mask);
 445         case PNFS_BLOCK_VOLUME_STRIPE:
 446                 return bl_parse_stripe(server, d, volumes, idx, gfp_mask);
 447         case PNFS_BLOCK_VOLUME_SCSI:
 448                 return bl_parse_scsi(server, d, volumes, idx, gfp_mask);
 449         default:
 450                 dprintk("unsupported volume type: %d\n", volumes[idx].type);
 451                 return -EIO;
 452         }
 453 }
 454
 455 struct nfs4_deviceid_node *
 456 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
 457                 gfp_t gfp_mask)
 458 {
 459         struct nfs4_deviceid_node *node = NULL;
 460         struct pnfs_block_volume *volumes;
 461         struct pnfs_block_dev *top;
 462         struct xdr_stream xdr;
 463         struct xdr_buf buf;
 464         struct page *scratch;
 465         int nr_volumes, ret, i;
 466         __be32 *p;
 467
 468         scratch = alloc_page(gfp_mask);
 469         if (!scratch)
 470                 goto out;
 471
 472         xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
 473         xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
 474
 475         p = xdr_inline_decode(&xdr, sizeof(__be32));
 476         if (!p)
 477                 goto out_free_scratch;
 478         nr_volumes = be32_to_cpup(p++);
 479
 480         volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
 481                           gfp_mask);
 482         if (!volumes)
 483                 goto out_free_scratch;
 484
 485         for (i = 0; i < nr_volumes; i++) {
 486                 ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
 487                 if (ret < 0)
 488                         goto out_free_volumes;
 489         }
 490
 491         top = kzalloc(sizeof(*top), gfp_mask);
 492         if (!top)
 493                 goto out_free_volumes;
 494
 495         ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
 496         if (ret) {
 497                 bl_free_device(top);
 498                 kfree(top);
 499                 goto out_free_volumes;
 500         }
 501
 502         node = &top->node;
 503         nfs4_init_deviceid_node(node, server, &pdev->dev_id);
 504
 505 out_free_volumes:
 506         kfree(volumes);
 507 out_free_scratch:
 508         __free_page(scratch);
 509 out:
 510         return node;
 511 }