2 * Copyright (c) 2014 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>
10 #include "blocklayout.h"
12 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
15 bl_free_device(struct pnfs_block_dev *dev)
17 if (dev->nr_children) {
20 for (i = 0; i < dev->nr_children; i++)
21 bl_free_device(&dev->children[i]);
25 blkdev_put(dev->bdev, FMODE_READ);
30 bl_free_deviceid_node(struct nfs4_deviceid_node *d)
32 struct pnfs_block_dev *dev =
33 container_of(d, struct pnfs_block_dev, node);
36 kfree_rcu(dev, node.rcu);
40 nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b)
45 p = xdr_inline_decode(xdr, 4);
48 b->type = be32_to_cpup(p++);
51 case PNFS_BLOCK_VOLUME_SIMPLE:
52 p = xdr_inline_decode(xdr, 4);
55 b->simple.nr_sigs = be32_to_cpup(p++);
56 if (!b->simple.nr_sigs) {
57 dprintk("no signature\n");
61 b->simple.len = 4 + 4;
62 for (i = 0; i < b->simple.nr_sigs; i++) {
63 p = xdr_inline_decode(xdr, 8 + 4);
66 p = xdr_decode_hyper(p, &b->simple.sigs[i].offset);
67 b->simple.sigs[i].sig_len = be32_to_cpup(p++);
69 p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
72 memcpy(&b->simple.sigs[i].sig, p,
73 b->simple.sigs[i].sig_len);
75 b->simple.len += 8 + 4 + b->simple.sigs[i].sig_len;
78 case PNFS_BLOCK_VOLUME_SLICE:
79 p = xdr_inline_decode(xdr, 8 + 8 + 4);
82 p = xdr_decode_hyper(p, &b->slice.start);
83 p = xdr_decode_hyper(p, &b->slice.len);
84 b->slice.volume = be32_to_cpup(p++);
86 case PNFS_BLOCK_VOLUME_CONCAT:
87 p = xdr_inline_decode(xdr, 4);
90 b->concat.volumes_count = be32_to_cpup(p++);
92 p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
95 for (i = 0; i < b->concat.volumes_count; i++)
96 b->concat.volumes[i] = be32_to_cpup(p++);
98 case PNFS_BLOCK_VOLUME_STRIPE:
99 p = xdr_inline_decode(xdr, 8 + 4);
102 p = xdr_decode_hyper(p, &b->stripe.chunk_size);
103 b->stripe.volumes_count = be32_to_cpup(p++);
105 p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
108 for (i = 0; i < b->stripe.volumes_count; i++)
109 b->stripe.volumes[i] = be32_to_cpup(p++);
112 dprintk("unknown volume type!\n");
119 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
120 struct pnfs_block_dev_map *map)
122 map->start = dev->start;
124 map->disk_offset = dev->disk_offset;
125 map->bdev = dev->bdev;
129 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
130 struct pnfs_block_dev_map *map)
134 for (i = 0; i < dev->nr_children; i++) {
135 struct pnfs_block_dev *child = &dev->children[i];
137 if (child->start > offset ||
138 child->start + child->len <= offset)
141 child->map(child, offset - child->start, map);
145 dprintk("%s: ran off loop!\n", __func__);
149 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
150 struct pnfs_block_dev_map *map)
152 struct pnfs_block_dev *child;
157 chunk = div_u64(offset, dev->chunk_size);
158 div_u64_rem(chunk, dev->nr_children, &chunk_idx);
160 if (chunk_idx > dev->nr_children) {
161 dprintk("%s: invalid chunk idx %d (%lld/%lld)\n",
162 __func__, chunk_idx, offset, dev->chunk_size);
163 /* error, should not happen */
167 /* truncate offset to the beginning of the stripe */
168 offset = chunk * dev->chunk_size;
170 /* disk offset of the stripe */
171 disk_offset = div_u64(offset, dev->nr_children);
173 child = &dev->children[chunk_idx];
174 child->map(child, disk_offset, map);
176 map->start += offset;
177 map->disk_offset += disk_offset;
178 map->len = dev->chunk_size;
183 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
184 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
188 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
189 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
191 struct pnfs_block_volume *v = &volumes[idx];
194 dev = bl_resolve_deviceid(server, v, gfp_mask);
198 d->bdev = blkdev_get_by_dev(dev, FMODE_READ, NULL);
199 if (IS_ERR(d->bdev)) {
200 printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
201 MAJOR(dev), MINOR(dev), PTR_ERR(d->bdev));
202 return PTR_ERR(d->bdev);
206 d->len = i_size_read(d->bdev->bd_inode);
207 d->map = bl_map_simple;
209 printk(KERN_INFO "pNFS: using block device %s\n",
210 d->bdev->bd_disk->disk_name);
215 bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d,
216 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
218 struct pnfs_block_volume *v = &volumes[idx];
221 ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
225 d->disk_offset = v->slice.start;
226 d->len = v->slice.len;
231 bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d,
232 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
234 struct pnfs_block_volume *v = &volumes[idx];
238 d->children = kcalloc(v->concat.volumes_count,
239 sizeof(struct pnfs_block_dev), GFP_KERNEL);
243 for (i = 0; i < v->concat.volumes_count; i++) {
244 ret = bl_parse_deviceid(server, &d->children[i],
245 volumes, v->concat.volumes[i], gfp_mask);
250 d->children[i].start += len;
251 len += d->children[i].len;
255 d->map = bl_map_concat;
260 bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d,
261 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
263 struct pnfs_block_volume *v = &volumes[idx];
267 d->children = kcalloc(v->stripe.volumes_count,
268 sizeof(struct pnfs_block_dev), GFP_KERNEL);
272 for (i = 0; i < v->stripe.volumes_count; i++) {
273 ret = bl_parse_deviceid(server, &d->children[i],
274 volumes, v->stripe.volumes[i], gfp_mask);
279 len += d->children[i].len;
283 d->chunk_size = v->stripe.chunk_size;
284 d->map = bl_map_stripe;
289 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
290 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
292 switch (volumes[idx].type) {
293 case PNFS_BLOCK_VOLUME_SIMPLE:
294 return bl_parse_simple(server, d, volumes, idx, gfp_mask);
295 case PNFS_BLOCK_VOLUME_SLICE:
296 return bl_parse_slice(server, d, volumes, idx, gfp_mask);
297 case PNFS_BLOCK_VOLUME_CONCAT:
298 return bl_parse_concat(server, d, volumes, idx, gfp_mask);
299 case PNFS_BLOCK_VOLUME_STRIPE:
300 return bl_parse_stripe(server, d, volumes, idx, gfp_mask);
302 dprintk("unsupported volume type: %d\n", volumes[idx].type);
307 struct nfs4_deviceid_node *
308 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
311 struct nfs4_deviceid_node *node = NULL;
312 struct pnfs_block_volume *volumes;
313 struct pnfs_block_dev *top;
314 struct xdr_stream xdr;
316 struct page *scratch;
317 int nr_volumes, ret, i;
320 scratch = alloc_page(gfp_mask);
324 xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
325 xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
327 p = xdr_inline_decode(&xdr, sizeof(__be32));
329 goto out_free_scratch;
330 nr_volumes = be32_to_cpup(p++);
332 volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
335 goto out_free_scratch;
337 for (i = 0; i < nr_volumes; i++) {
338 ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
340 goto out_free_volumes;
343 top = kzalloc(sizeof(*top), gfp_mask);
345 goto out_free_volumes;
347 ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
351 goto out_free_volumes;
355 nfs4_init_deviceid_node(node, server, &pdev->dev_id);
360 __free_page(scratch);