2 * Copyright (c) 2014-2016 Christoph Hellwig.
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>
11 #include "blocklayout.h"
13 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
16 bl_free_device(struct pnfs_block_dev *dev)
18 if (dev->nr_children) {
21 for (i = 0; i < dev->nr_children; i++)
22 bl_free_device(&dev->children[i]);
25 if (dev->pr_registered) {
26 const struct pr_ops *ops =
27 dev->bdev->bd_disk->fops->pr_ops;
30 error = ops->pr_register(dev->bdev, dev->pr_key, 0,
33 pr_err("failed to unregister PR key.\n");
37 blkdev_put(dev->bdev, FMODE_READ | FMODE_WRITE);
42 bl_free_deviceid_node(struct nfs4_deviceid_node *d)
44 struct pnfs_block_dev *dev =
45 container_of(d, struct pnfs_block_dev, node);
48 kfree_rcu(dev, node.rcu);
52 nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b)
57 p = xdr_inline_decode(xdr, 4);
60 b->type = be32_to_cpup(p++);
63 case PNFS_BLOCK_VOLUME_SIMPLE:
64 p = xdr_inline_decode(xdr, 4);
67 b->simple.nr_sigs = be32_to_cpup(p++);
68 if (!b->simple.nr_sigs || b->simple.nr_sigs > PNFS_BLOCK_MAX_UUIDS) {
69 dprintk("Bad signature count: %d\n", b->simple.nr_sigs);
73 b->simple.len = 4 + 4;
74 for (i = 0; i < b->simple.nr_sigs; i++) {
75 p = xdr_inline_decode(xdr, 8 + 4);
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);
86 p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
89 memcpy(&b->simple.sigs[i].sig, p,
90 b->simple.sigs[i].sig_len);
92 b->simple.len += 8 + 4 + \
93 (XDR_QUADLEN(b->simple.sigs[i].sig_len) << 2);
96 case PNFS_BLOCK_VOLUME_SLICE:
97 p = xdr_inline_decode(xdr, 8 + 8 + 4);
100 p = xdr_decode_hyper(p, &b->slice.start);
101 p = xdr_decode_hyper(p, &b->slice.len);
102 b->slice.volume = be32_to_cpup(p++);
104 case PNFS_BLOCK_VOLUME_CONCAT:
105 p = xdr_inline_decode(xdr, 4);
109 b->concat.volumes_count = be32_to_cpup(p++);
110 if (b->concat.volumes_count > PNFS_BLOCK_MAX_DEVICES) {
111 dprintk("Too many volumes: %d\n", b->concat.volumes_count);
115 p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
118 for (i = 0; i < b->concat.volumes_count; i++)
119 b->concat.volumes[i] = be32_to_cpup(p++);
121 case PNFS_BLOCK_VOLUME_STRIPE:
122 p = xdr_inline_decode(xdr, 8 + 4);
126 p = xdr_decode_hyper(p, &b->stripe.chunk_size);
127 b->stripe.volumes_count = be32_to_cpup(p++);
128 if (b->stripe.volumes_count > PNFS_BLOCK_MAX_DEVICES) {
129 dprintk("Too many volumes: %d\n", b->stripe.volumes_count);
133 p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
136 for (i = 0; i < b->stripe.volumes_count; i++)
137 b->stripe.volumes[i] = be32_to_cpup(p++);
139 case PNFS_BLOCK_VOLUME_SCSI:
140 p = xdr_inline_decode(xdr, 4 + 4 + 4);
143 b->scsi.code_set = be32_to_cpup(p++);
144 b->scsi.designator_type = be32_to_cpup(p++);
145 b->scsi.designator_len = be32_to_cpup(p++);
146 p = xdr_inline_decode(xdr, b->scsi.designator_len);
149 if (b->scsi.designator_len > 256)
151 memcpy(&b->scsi.designator, p, b->scsi.designator_len);
152 p = xdr_inline_decode(xdr, 8);
155 p = xdr_decode_hyper(p, &b->scsi.pr_key);
158 dprintk("unknown volume type!\n");
165 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
166 struct pnfs_block_dev_map *map)
168 map->start = dev->start;
170 map->disk_offset = dev->disk_offset;
171 map->bdev = dev->bdev;
175 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
176 struct pnfs_block_dev_map *map)
180 for (i = 0; i < dev->nr_children; i++) {
181 struct pnfs_block_dev *child = &dev->children[i];
183 if (child->start > offset ||
184 child->start + child->len <= offset)
187 child->map(child, offset - child->start, map);
191 dprintk("%s: ran off loop!\n", __func__);
195 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
196 struct pnfs_block_dev_map *map)
198 struct pnfs_block_dev *child;
203 chunk = div_u64(offset, dev->chunk_size);
204 div_u64_rem(chunk, dev->nr_children, &chunk_idx);
206 if (chunk_idx >= dev->nr_children) {
207 dprintk("%s: invalid chunk idx %d (%lld/%lld)\n",
208 __func__, chunk_idx, offset, dev->chunk_size);
209 /* error, should not happen */
213 /* truncate offset to the beginning of the stripe */
214 offset = chunk * dev->chunk_size;
216 /* disk offset of the stripe */
217 disk_offset = div_u64(offset, dev->nr_children);
219 child = &dev->children[chunk_idx];
220 child->map(child, disk_offset, map);
222 map->start += offset;
223 map->disk_offset += disk_offset;
224 map->len = dev->chunk_size;
229 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
230 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
234 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
235 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
237 struct pnfs_block_volume *v = &volumes[idx];
238 struct block_device *bdev;
241 dev = bl_resolve_deviceid(server, v, gfp_mask);
245 bdev = blkdev_get_by_dev(dev, FMODE_READ | FMODE_WRITE, NULL);
247 printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
248 MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
249 return PTR_ERR(bdev);
254 d->len = i_size_read(d->bdev->bd_inode);
255 d->map = bl_map_simple;
257 printk(KERN_INFO "pNFS: using block device %s\n",
258 d->bdev->bd_disk->disk_name);
263 bl_validate_designator(struct pnfs_block_volume *v)
265 switch (v->scsi.designator_type) {
266 case PS_DESIGNATOR_EUI64:
267 if (v->scsi.code_set != PS_CODE_SET_BINARY)
270 if (v->scsi.designator_len != 8 &&
271 v->scsi.designator_len != 10 &&
272 v->scsi.designator_len != 16)
276 case PS_DESIGNATOR_NAA:
277 if (v->scsi.code_set != PS_CODE_SET_BINARY)
280 if (v->scsi.designator_len != 8 &&
281 v->scsi.designator_len != 16)
285 case PS_DESIGNATOR_T10:
286 case PS_DESIGNATOR_NAME:
287 pr_err("pNFS: unsupported designator "
288 "(code set %d, type %d, len %d.\n",
290 v->scsi.designator_type,
291 v->scsi.designator_len);
294 pr_err("pNFS: invalid designator "
295 "(code set %d, type %d, len %d.\n",
297 v->scsi.designator_type,
298 v->scsi.designator_len);
304 * Try to open the udev path for the WWN. At least on Debian the udev
305 * by-id path will always point to the dm-multipath device if one exists.
307 static struct block_device *
308 bl_open_udev_path(struct pnfs_block_volume *v)
310 struct block_device *bdev;
313 devname = kasprintf(GFP_KERNEL, "/dev/disk/by-id/wwn-0x%*phN",
314 v->scsi.designator_len, v->scsi.designator);
316 return ERR_PTR(-ENOMEM);
318 bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
320 pr_warn("pNFS: failed to open device %s (%ld)\n",
321 devname, PTR_ERR(bdev));
329 * Try to open the RH/Fedora specific dm-mpath udev path for this WWN, as the
330 * wwn- links will only point to the first discovered SCSI device there.
332 static struct block_device *
333 bl_open_dm_mpath_udev_path(struct pnfs_block_volume *v)
335 struct block_device *bdev;
338 devname = kasprintf(GFP_KERNEL,
339 "/dev/disk/by-id/dm-uuid-mpath-%d%*phN",
340 v->scsi.designator_type,
341 v->scsi.designator_len, v->scsi.designator);
343 return ERR_PTR(-ENOMEM);
345 bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
351 bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d,
352 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
354 struct pnfs_block_volume *v = &volumes[idx];
355 struct block_device *bdev;
356 const struct pr_ops *ops;
359 if (!bl_validate_designator(v))
362 bdev = bl_open_dm_mpath_udev_path(v);
364 bdev = bl_open_udev_path(v);
366 return PTR_ERR(bdev);
369 d->len = i_size_read(d->bdev->bd_inode);
370 d->map = bl_map_simple;
371 d->pr_key = v->scsi.pr_key;
373 pr_info("pNFS: using block device %s (reservation key 0x%llx)\n",
374 d->bdev->bd_disk->disk_name, d->pr_key);
376 ops = d->bdev->bd_disk->fops->pr_ops;
378 pr_err("pNFS: block device %s does not support reservations.",
379 d->bdev->bd_disk->disk_name);
384 error = ops->pr_register(d->bdev, 0, d->pr_key, true);
386 pr_err("pNFS: failed to register key for block device %s.",
387 d->bdev->bd_disk->disk_name);
391 d->pr_registered = true;
395 blkdev_put(d->bdev, FMODE_READ | FMODE_WRITE);
400 bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d,
401 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
403 struct pnfs_block_volume *v = &volumes[idx];
406 ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
410 d->disk_offset = v->slice.start;
411 d->len = v->slice.len;
416 bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d,
417 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
419 struct pnfs_block_volume *v = &volumes[idx];
423 d->children = kcalloc(v->concat.volumes_count,
424 sizeof(struct pnfs_block_dev), GFP_KERNEL);
428 for (i = 0; i < v->concat.volumes_count; i++) {
429 ret = bl_parse_deviceid(server, &d->children[i],
430 volumes, v->concat.volumes[i], gfp_mask);
435 d->children[i].start += len;
436 len += d->children[i].len;
440 d->map = bl_map_concat;
445 bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d,
446 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
448 struct pnfs_block_volume *v = &volumes[idx];
452 d->children = kcalloc(v->stripe.volumes_count,
453 sizeof(struct pnfs_block_dev), GFP_KERNEL);
457 for (i = 0; i < v->stripe.volumes_count; i++) {
458 ret = bl_parse_deviceid(server, &d->children[i],
459 volumes, v->stripe.volumes[i], gfp_mask);
464 len += d->children[i].len;
468 d->chunk_size = v->stripe.chunk_size;
469 d->map = bl_map_stripe;
474 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
475 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
477 switch (volumes[idx].type) {
478 case PNFS_BLOCK_VOLUME_SIMPLE:
479 return bl_parse_simple(server, d, volumes, idx, gfp_mask);
480 case PNFS_BLOCK_VOLUME_SLICE:
481 return bl_parse_slice(server, d, volumes, idx, gfp_mask);
482 case PNFS_BLOCK_VOLUME_CONCAT:
483 return bl_parse_concat(server, d, volumes, idx, gfp_mask);
484 case PNFS_BLOCK_VOLUME_STRIPE:
485 return bl_parse_stripe(server, d, volumes, idx, gfp_mask);
486 case PNFS_BLOCK_VOLUME_SCSI:
487 return bl_parse_scsi(server, d, volumes, idx, gfp_mask);
489 dprintk("unsupported volume type: %d\n", volumes[idx].type);
494 struct nfs4_deviceid_node *
495 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
498 struct nfs4_deviceid_node *node = NULL;
499 struct pnfs_block_volume *volumes;
500 struct pnfs_block_dev *top;
501 struct xdr_stream xdr;
503 struct page *scratch;
504 int nr_volumes, ret, i;
507 scratch = alloc_page(gfp_mask);
511 xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
512 xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
514 p = xdr_inline_decode(&xdr, sizeof(__be32));
516 goto out_free_scratch;
517 nr_volumes = be32_to_cpup(p++);
519 volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
522 goto out_free_scratch;
524 for (i = 0; i < nr_volumes; i++) {
525 ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
527 goto out_free_volumes;
530 top = kzalloc(sizeof(*top), gfp_mask);
532 goto out_free_volumes;
534 ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
538 goto out_free_volumes;
542 nfs4_init_deviceid_node(node, server, &pdev->dev_id);
547 __free_page(scratch);