nfs/blocklayout/dev.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Copyright (c) 2014-2016 Christoph Hellwig.
0004  */
0005 #include <linux/sunrpc/svc.h>
0006 #include <linux/blkdev.h>
0007 #include <linux/nfs4.h>
0008 #include <linux/nfs_fs.h>
0009 #include <linux/nfs_xdr.h>
0010 #include <linux/pr.h>
0011
0012 #include "blocklayout.h"
0013
0014 #define NFSDBG_FACILITY     NFSDBG_PNFS_LD
0015
0016 static void
0017 bl_free_device(struct pnfs_block_dev *dev)
0018 {
0019     if (dev->nr_children) {
0020         int i;
0021
0022         for (i = 0; i < dev->nr_children; i++)
0023             bl_free_device(&dev->children[i]);
0024         kfree(dev->children);
0025     } else {
0026         if (dev->pr_registered) {
0027             const struct pr_ops *ops =
0028                 dev->bdev->bd_disk->fops->pr_ops;
0029             int error;
0030
0031             error = ops->pr_register(dev->bdev, dev->pr_key, 0,
0032                 false);
0033             if (error)
0034                 pr_err("failed to unregister PR key.\n");
0035         }
0036
0037         if (dev->bdev)
0038             blkdev_put(dev->bdev, FMODE_READ | FMODE_WRITE);
0039     }
0040 }
0041
0042 void
0043 bl_free_deviceid_node(struct nfs4_deviceid_node *d)
0044 {
0045     struct pnfs_block_dev *dev =
0046         container_of(d, struct pnfs_block_dev, node);
0047
0048     bl_free_device(dev);
0049     kfree_rcu(dev, node.rcu);
0050 }
0051
0052 static int
0053 nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b)
0054 {
0055     __be32 *p;
0056     int i;
0057
0058     p = xdr_inline_decode(xdr, 4);
0059     if (!p)
0060         return -EIO;
0061     b->type = be32_to_cpup(p++);
0062
0063     switch (b->type) {
0064     case PNFS_BLOCK_VOLUME_SIMPLE:
0065         p = xdr_inline_decode(xdr, 4);
0066         if (!p)
0067             return -EIO;
0068         b->simple.nr_sigs = be32_to_cpup(p++);
0069         if (!b->simple.nr_sigs || b->simple.nr_sigs > PNFS_BLOCK_MAX_UUIDS) {
0070             dprintk("Bad signature count: %d\n", b->simple.nr_sigs);
0071             return -EIO;
0072         }
0073
0074         b->simple.len = 4 + 4;
0075         for (i = 0; i < b->simple.nr_sigs; i++) {
0076             p = xdr_inline_decode(xdr, 8 + 4);
0077             if (!p)
0078                 return -EIO;
0079             p = xdr_decode_hyper(p, &b->simple.sigs[i].offset);
0080             b->simple.sigs[i].sig_len = be32_to_cpup(p++);
0081             if (b->simple.sigs[i].sig_len > PNFS_BLOCK_UUID_LEN) {
0082                 pr_info("signature too long: %d\n",
0083                     b->simple.sigs[i].sig_len);
0084                 return -EIO;
0085             }
0086
0087             p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
0088             if (!p)
0089                 return -EIO;
0090             memcpy(&b->simple.sigs[i].sig, p,
0091                 b->simple.sigs[i].sig_len);
0092
0093             b->simple.len += 8 + 4 + \
0094                 (XDR_QUADLEN(b->simple.sigs[i].sig_len) << 2);
0095         }
0096         break;
0097     case PNFS_BLOCK_VOLUME_SLICE:
0098         p = xdr_inline_decode(xdr, 8 + 8 + 4);
0099         if (!p)
0100             return -EIO;
0101         p = xdr_decode_hyper(p, &b->slice.start);
0102         p = xdr_decode_hyper(p, &b->slice.len);
0103         b->slice.volume = be32_to_cpup(p++);
0104         break;
0105     case PNFS_BLOCK_VOLUME_CONCAT:
0106         p = xdr_inline_decode(xdr, 4);
0107         if (!p)
0108             return -EIO;
0109
0110         b->concat.volumes_count = be32_to_cpup(p++);
0111         if (b->concat.volumes_count > PNFS_BLOCK_MAX_DEVICES) {
0112             dprintk("Too many volumes: %d\n", b->concat.volumes_count);
0113             return -EIO;
0114         }
0115
0116         p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
0117         if (!p)
0118             return -EIO;
0119         for (i = 0; i < b->concat.volumes_count; i++)
0120             b->concat.volumes[i] = be32_to_cpup(p++);
0121         break;
0122     case PNFS_BLOCK_VOLUME_STRIPE:
0123         p = xdr_inline_decode(xdr, 8 + 4);
0124         if (!p)
0125             return -EIO;
0126
0127         p = xdr_decode_hyper(p, &b->stripe.chunk_size);
0128         b->stripe.volumes_count = be32_to_cpup(p++);
0129         if (b->stripe.volumes_count > PNFS_BLOCK_MAX_DEVICES) {
0130             dprintk("Too many volumes: %d\n", b->stripe.volumes_count);
0131             return -EIO;
0132         }
0133
0134         p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
0135         if (!p)
0136             return -EIO;
0137         for (i = 0; i < b->stripe.volumes_count; i++)
0138             b->stripe.volumes[i] = be32_to_cpup(p++);
0139         break;
0140     case PNFS_BLOCK_VOLUME_SCSI:
0141         p = xdr_inline_decode(xdr, 4 + 4 + 4);
0142         if (!p)
0143             return -EIO;
0144         b->scsi.code_set = be32_to_cpup(p++);
0145         b->scsi.designator_type = be32_to_cpup(p++);
0146         b->scsi.designator_len = be32_to_cpup(p++);
0147         p = xdr_inline_decode(xdr, b->scsi.designator_len);
0148         if (!p)
0149             return -EIO;
0150         if (b->scsi.designator_len > 256)
0151             return -EIO;
0152         memcpy(&b->scsi.designator, p, b->scsi.designator_len);
0153         p = xdr_inline_decode(xdr, 8);
0154         if (!p)
0155             return -EIO;
0156         p = xdr_decode_hyper(p, &b->scsi.pr_key);
0157         break;
0158     default:
0159         dprintk("unknown volume type!\n");
0160         return -EIO;
0161     }
0162
0163     return 0;
0164 }
0165
0166 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
0167         struct pnfs_block_dev_map *map)
0168 {
0169     map->start = dev->start;
0170     map->len = dev->len;
0171     map->disk_offset = dev->disk_offset;
0172     map->bdev = dev->bdev;
0173     return true;
0174 }
0175
0176 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
0177         struct pnfs_block_dev_map *map)
0178 {
0179     int i;
0180
0181     for (i = 0; i < dev->nr_children; i++) {
0182         struct pnfs_block_dev *child = &dev->children[i];
0183
0184         if (child->start > offset ||
0185             child->start + child->len <= offset)
0186             continue;
0187
0188         child->map(child, offset - child->start, map);
0189         return true;
0190     }
0191
0192     dprintk("%s: ran off loop!\n", __func__);
0193     return false;
0194 }
0195
0196 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
0197         struct pnfs_block_dev_map *map)
0198 {
0199     struct pnfs_block_dev *child;
0200     u64 chunk;
0201     u32 chunk_idx;
0202     u64 disk_offset;
0203
0204     chunk = div_u64(offset, dev->chunk_size);
0205     div_u64_rem(chunk, dev->nr_children, &chunk_idx);
0206
0207     if (chunk_idx >= dev->nr_children) {
0208         dprintk("%s: invalid chunk idx %d (%lld/%lld)\n",
0209             __func__, chunk_idx, offset, dev->chunk_size);
0210         /* error, should not happen */
0211         return false;
0212     }
0213
0214     /* truncate offset to the beginning of the stripe */
0215     offset = chunk * dev->chunk_size;
0216
0217     /* disk offset of the stripe */
0218     disk_offset = div_u64(offset, dev->nr_children);
0219
0220     child = &dev->children[chunk_idx];
0221     child->map(child, disk_offset, map);
0222
0223     map->start += offset;
0224     map->disk_offset += disk_offset;
0225     map->len = dev->chunk_size;
0226     return true;
0227 }
0228
0229 static int
0230 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
0231         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
0232
0233
0234 static int
0235 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
0236         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
0237 {
0238     struct pnfs_block_volume *v = &volumes[idx];
0239     struct block_device *bdev;
0240     dev_t dev;
0241
0242     dev = bl_resolve_deviceid(server, v, gfp_mask);
0243     if (!dev)
0244         return -EIO;
0245
0246     bdev = blkdev_get_by_dev(dev, FMODE_READ | FMODE_WRITE, NULL);
0247     if (IS_ERR(bdev)) {
0248         printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
0249             MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
0250         return PTR_ERR(bdev);
0251     }
0252     d->bdev = bdev;
0253
0254
0255     d->len = bdev_nr_bytes(d->bdev);
0256     d->map = bl_map_simple;
0257
0258     printk(KERN_INFO "pNFS: using block device %s\n",
0259         d->bdev->bd_disk->disk_name);
0260     return 0;
0261 }
0262
0263 static bool
0264 bl_validate_designator(struct pnfs_block_volume *v)
0265 {
0266     switch (v->scsi.designator_type) {
0267     case PS_DESIGNATOR_EUI64:
0268         if (v->scsi.code_set != PS_CODE_SET_BINARY)
0269             return false;
0270
0271         if (v->scsi.designator_len != 8 &&
0272             v->scsi.designator_len != 10 &&
0273             v->scsi.designator_len != 16)
0274             return false;
0275
0276         return true;
0277     case PS_DESIGNATOR_NAA:
0278         if (v->scsi.code_set != PS_CODE_SET_BINARY)
0279             return false;
0280
0281         if (v->scsi.designator_len != 8 &&
0282             v->scsi.designator_len != 16)
0283             return false;
0284
0285         return true;
0286     case PS_DESIGNATOR_T10:
0287     case PS_DESIGNATOR_NAME:
0288         pr_err("pNFS: unsupported designator "
0289             "(code set %d, type %d, len %d.\n",
0290             v->scsi.code_set,
0291             v->scsi.designator_type,
0292             v->scsi.designator_len);
0293         return false;
0294     default:
0295         pr_err("pNFS: invalid designator "
0296             "(code set %d, type %d, len %d.\n",
0297             v->scsi.code_set,
0298             v->scsi.designator_type,
0299             v->scsi.designator_len);
0300         return false;
0301     }
0302 }
0303
0304 static struct block_device *
0305 bl_open_path(struct pnfs_block_volume *v, const char *prefix)
0306 {
0307     struct block_device *bdev;
0308     const char *devname;
0309
0310     devname = kasprintf(GFP_KERNEL, "/dev/disk/by-id/%s%*phN",
0311             prefix, v->scsi.designator_len, v->scsi.designator);
0312     if (!devname)
0313         return ERR_PTR(-ENOMEM);
0314
0315     bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
0316     if (IS_ERR(bdev)) {
0317         pr_warn("pNFS: failed to open device %s (%ld)\n",
0318             devname, PTR_ERR(bdev));
0319     }
0320
0321     kfree(devname);
0322     return bdev;
0323 }
0324
0325 static int
0326 bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d,
0327         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
0328 {
0329     struct pnfs_block_volume *v = &volumes[idx];
0330     struct block_device *bdev;
0331     const struct pr_ops *ops;
0332     int error;
0333
0334     if (!bl_validate_designator(v))
0335         return -EINVAL;
0336
0337     /*
0338      * Try to open the RH/Fedora specific dm-mpath udev path first, as the
0339      * wwn- links will only point to the first discovered SCSI device there.
0340      * On other distributions like Debian, the default SCSI by-id path will
0341      * point to the dm-multipath device if one exists.
0342      */
0343     bdev = bl_open_path(v, "dm-uuid-mpath-0x");
0344     if (IS_ERR(bdev))
0345         bdev = bl_open_path(v, "wwn-0x");
0346     if (IS_ERR(bdev))
0347         return PTR_ERR(bdev);
0348     d->bdev = bdev;
0349
0350     d->len = bdev_nr_bytes(d->bdev);
0351     d->map = bl_map_simple;
0352     d->pr_key = v->scsi.pr_key;
0353
0354     pr_info("pNFS: using block device %s (reservation key 0x%llx)\n",
0355         d->bdev->bd_disk->disk_name, d->pr_key);
0356
0357     ops = d->bdev->bd_disk->fops->pr_ops;
0358     if (!ops) {
0359         pr_err("pNFS: block device %s does not support reservations.",
0360                 d->bdev->bd_disk->disk_name);
0361         error = -EINVAL;
0362         goto out_blkdev_put;
0363     }
0364
0365     error = ops->pr_register(d->bdev, 0, d->pr_key, true);
0366     if (error) {
0367         pr_err("pNFS: failed to register key for block device %s.",
0368                 d->bdev->bd_disk->disk_name);
0369         goto out_blkdev_put;
0370     }
0371
0372     d->pr_registered = true;
0373     return 0;
0374
0375 out_blkdev_put:
0376     blkdev_put(d->bdev, FMODE_READ | FMODE_WRITE);
0377     return error;
0378 }
0379
0380 static int
0381 bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d,
0382         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
0383 {
0384     struct pnfs_block_volume *v = &volumes[idx];
0385     int ret;
0386
0387     ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
0388     if (ret)
0389         return ret;
0390
0391     d->disk_offset = v->slice.start;
0392     d->len = v->slice.len;
0393     return 0;
0394 }
0395
0396 static int
0397 bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d,
0398         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
0399 {
0400     struct pnfs_block_volume *v = &volumes[idx];
0401     u64 len = 0;
0402     int ret, i;
0403
0404     d->children = kcalloc(v->concat.volumes_count,
0405             sizeof(struct pnfs_block_dev), GFP_KERNEL);
0406     if (!d->children)
0407         return -ENOMEM;
0408
0409     for (i = 0; i < v->concat.volumes_count; i++) {
0410         ret = bl_parse_deviceid(server, &d->children[i],
0411                 volumes, v->concat.volumes[i], gfp_mask);
0412         if (ret)
0413             return ret;
0414
0415         d->nr_children++;
0416         d->children[i].start += len;
0417         len += d->children[i].len;
0418     }
0419
0420     d->len = len;
0421     d->map = bl_map_concat;
0422     return 0;
0423 }
0424
0425 static int
0426 bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d,
0427         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
0428 {
0429     struct pnfs_block_volume *v = &volumes[idx];
0430     u64 len = 0;
0431     int ret, i;
0432
0433     d->children = kcalloc(v->stripe.volumes_count,
0434             sizeof(struct pnfs_block_dev), GFP_KERNEL);
0435     if (!d->children)
0436         return -ENOMEM;
0437
0438     for (i = 0; i < v->stripe.volumes_count; i++) {
0439         ret = bl_parse_deviceid(server, &d->children[i],
0440                 volumes, v->stripe.volumes[i], gfp_mask);
0441         if (ret)
0442             return ret;
0443
0444         d->nr_children++;
0445         len += d->children[i].len;
0446     }
0447
0448     d->len = len;
0449     d->chunk_size = v->stripe.chunk_size;
0450     d->map = bl_map_stripe;
0451     return 0;
0452 }
0453
0454 static int
0455 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
0456         struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
0457 {
0458     switch (volumes[idx].type) {
0459     case PNFS_BLOCK_VOLUME_SIMPLE:
0460         return bl_parse_simple(server, d, volumes, idx, gfp_mask);
0461     case PNFS_BLOCK_VOLUME_SLICE:
0462         return bl_parse_slice(server, d, volumes, idx, gfp_mask);
0463     case PNFS_BLOCK_VOLUME_CONCAT:
0464         return bl_parse_concat(server, d, volumes, idx, gfp_mask);
0465     case PNFS_BLOCK_VOLUME_STRIPE:
0466         return bl_parse_stripe(server, d, volumes, idx, gfp_mask);
0467     case PNFS_BLOCK_VOLUME_SCSI:
0468         return bl_parse_scsi(server, d, volumes, idx, gfp_mask);
0469     default:
0470         dprintk("unsupported volume type: %d\n", volumes[idx].type);
0471         return -EIO;
0472     }
0473 }
0474
0475 struct nfs4_deviceid_node *
0476 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
0477         gfp_t gfp_mask)
0478 {
0479     struct nfs4_deviceid_node *node = NULL;
0480     struct pnfs_block_volume *volumes;
0481     struct pnfs_block_dev *top;
0482     struct xdr_stream xdr;
0483     struct xdr_buf buf;
0484     struct page *scratch;
0485     int nr_volumes, ret, i;
0486     __be32 *p;
0487
0488     scratch = alloc_page(gfp_mask);
0489     if (!scratch)
0490         goto out;
0491
0492     xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
0493     xdr_set_scratch_page(&xdr, scratch);
0494
0495     p = xdr_inline_decode(&xdr, sizeof(__be32));
0496     if (!p)
0497         goto out_free_scratch;
0498     nr_volumes = be32_to_cpup(p++);
0499
0500     volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
0501               gfp_mask);
0502     if (!volumes)
0503         goto out_free_scratch;
0504
0505     for (i = 0; i < nr_volumes; i++) {
0506         ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
0507         if (ret < 0)
0508             goto out_free_volumes;
0509     }
0510
0511     top = kzalloc(sizeof(*top), gfp_mask);
0512     if (!top)
0513         goto out_free_volumes;
0514
0515     ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
0516
0517     node = &top->node;
0518     nfs4_init_deviceid_node(node, server, &pdev->dev_id);
0519     if (ret)
0520         nfs4_mark_deviceid_unavailable(node);
0521
0522 out_free_volumes:
0523     kfree(volumes);
0524 out_free_scratch:
0525     __free_page(scratch);
0526 out:
0527     return node;
0528 }