1/*
2 * Copyright (c) 2014 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
10#include "blocklayout.h"
11
12#define NFSDBG_FACILITY		NFSDBG_PNFS_LD
13
14static void
15bl_free_device(struct pnfs_block_dev *dev)
16{
17	if (dev->nr_children) {
18		int i;
19
20		for (i = 0; i < dev->nr_children; i++)
21			bl_free_device(&dev->children[i]);
22		kfree(dev->children);
23	} else {
24		if (dev->bdev)
25			blkdev_put(dev->bdev, FMODE_READ);
26	}
27}
28
29void
30bl_free_deviceid_node(struct nfs4_deviceid_node *d)
31{
32	struct pnfs_block_dev *dev =
33		container_of(d, struct pnfs_block_dev, node);
34
35	bl_free_device(dev);
36	kfree_rcu(dev, node.rcu);
37}
38
39static int
40nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b)
41{
42	__be32 *p;
43	int i;
44
45	p = xdr_inline_decode(xdr, 4);
46	if (!p)
47		return -EIO;
48	b->type = be32_to_cpup(p++);
49
50	switch (b->type) {
51	case PNFS_BLOCK_VOLUME_SIMPLE:
52		p = xdr_inline_decode(xdr, 4);
53		if (!p)
54			return -EIO;
55		b->simple.nr_sigs = be32_to_cpup(p++);
56		if (!b->simple.nr_sigs) {
57			dprintk("no signature\n");
58			return -EIO;
59		}
60
61		b->simple.len = 4 + 4;
62		for (i = 0; i < b->simple.nr_sigs; i++) {
63			p = xdr_inline_decode(xdr, 8 + 4);
64			if (!p)
65				return -EIO;
66			p = xdr_decode_hyper(p, &b->simple.sigs[i].offset);
67			b->simple.sigs[i].sig_len = be32_to_cpup(p++);
68
69			p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
70			if (!p)
71				return -EIO;
72			memcpy(&b->simple.sigs[i].sig, p,
73				b->simple.sigs[i].sig_len);
74
75			b->simple.len += 8 + 4 + b->simple.sigs[i].sig_len;
76		}
77		break;
78	case PNFS_BLOCK_VOLUME_SLICE:
79		p = xdr_inline_decode(xdr, 8 + 8 + 4);
80		if (!p)
81			return -EIO;
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++);
85		break;
86	case PNFS_BLOCK_VOLUME_CONCAT:
87		p = xdr_inline_decode(xdr, 4);
88		if (!p)
89			return -EIO;
90		b->concat.volumes_count = be32_to_cpup(p++);
91
92		p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
93		if (!p)
94			return -EIO;
95		for (i = 0; i < b->concat.volumes_count; i++)
96			b->concat.volumes[i] = be32_to_cpup(p++);
97		break;
98	case PNFS_BLOCK_VOLUME_STRIPE:
99		p = xdr_inline_decode(xdr, 8 + 4);
100		if (!p)
101			return -EIO;
102		p = xdr_decode_hyper(p, &b->stripe.chunk_size);
103		b->stripe.volumes_count = be32_to_cpup(p++);
104
105		p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
106		if (!p)
107			return -EIO;
108		for (i = 0; i < b->stripe.volumes_count; i++)
109			b->stripe.volumes[i] = be32_to_cpup(p++);
110		break;
111	default:
112		dprintk("unknown volume type!\n");
113		return -EIO;
114	}
115
116	return 0;
117}
118
119static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
120		struct pnfs_block_dev_map *map)
121{
122	map->start = dev->start;
123	map->len = dev->len;
124	map->disk_offset = dev->disk_offset;
125	map->bdev = dev->bdev;
126	return true;
127}
128
129static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
130		struct pnfs_block_dev_map *map)
131{
132	int i;
133
134	for (i = 0; i < dev->nr_children; i++) {
135		struct pnfs_block_dev *child = &dev->children[i];
136
137		if (child->start > offset ||
138		    child->start + child->len <= offset)
139			continue;
140
141		child->map(child, offset - child->start, map);
142		return true;
143	}
144
145	dprintk("%s: ran off loop!\n", __func__);
146	return false;
147}
148
149static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
150		struct pnfs_block_dev_map *map)
151{
152	struct pnfs_block_dev *child;
153	u64 chunk;
154	u32 chunk_idx;
155	u64 disk_offset;
156
157	chunk = div_u64(offset, dev->chunk_size);
158	div_u64_rem(chunk, dev->nr_children, &chunk_idx);
159
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 */
164		return false;
165	}
166
167	/* truncate offset to the beginning of the stripe */
168	offset = chunk * dev->chunk_size;
169
170	/* disk offset of the stripe */
171	disk_offset = div_u64(offset, dev->nr_children);
172
173	child = &dev->children[chunk_idx];
174	child->map(child, disk_offset, map);
175
176	map->start += offset;
177	map->disk_offset += disk_offset;
178	map->len = dev->chunk_size;
179	return true;
180}
181
182static int
183bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
184		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
185
186
187static int
188bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
189		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
190{
191	struct pnfs_block_volume *v = &volumes[idx];
192	dev_t dev;
193
194	dev = bl_resolve_deviceid(server, v, gfp_mask);
195	if (!dev)
196		return -EIO;
197
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);
203	}
204
205
206	d->len = i_size_read(d->bdev->bd_inode);
207	d->map = bl_map_simple;
208
209	printk(KERN_INFO "pNFS: using block device %s\n",
210		d->bdev->bd_disk->disk_name);
211	return 0;
212}
213
214static int
215bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d,
216		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
217{
218	struct pnfs_block_volume *v = &volumes[idx];
219	int ret;
220
221	ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
222	if (ret)
223		return ret;
224
225	d->disk_offset = v->slice.start;
226	d->len = v->slice.len;
227	return 0;
228}
229
230static int
231bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d,
232		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
233{
234	struct pnfs_block_volume *v = &volumes[idx];
235	u64 len = 0;
236	int ret, i;
237
238	d->children = kcalloc(v->concat.volumes_count,
239			sizeof(struct pnfs_block_dev), GFP_KERNEL);
240	if (!d->children)
241		return -ENOMEM;
242
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);
246		if (ret)
247			return ret;
248
249		d->nr_children++;
250		d->children[i].start += len;
251		len += d->children[i].len;
252	}
253
254	d->len = len;
255	d->map = bl_map_concat;
256	return 0;
257}
258
259static int
260bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d,
261		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
262{
263	struct pnfs_block_volume *v = &volumes[idx];
264	u64 len = 0;
265	int ret, i;
266
267	d->children = kcalloc(v->stripe.volumes_count,
268			sizeof(struct pnfs_block_dev), GFP_KERNEL);
269	if (!d->children)
270		return -ENOMEM;
271
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);
275		if (ret)
276			return ret;
277
278		d->nr_children++;
279		len += d->children[i].len;
280	}
281
282	d->len = len;
283	d->chunk_size = v->stripe.chunk_size;
284	d->map = bl_map_stripe;
285	return 0;
286}
287
288static int
289bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
290		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
291{
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);
301	default:
302		dprintk("unsupported volume type: %d\n", volumes[idx].type);
303		return -EIO;
304	}
305}
306
307struct nfs4_deviceid_node *
308bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
309		gfp_t gfp_mask)
310{
311	struct nfs4_deviceid_node *node = NULL;
312	struct pnfs_block_volume *volumes;
313	struct pnfs_block_dev *top;
314	struct xdr_stream xdr;
315	struct xdr_buf buf;
316	struct page *scratch;
317	int nr_volumes, ret, i;
318	__be32 *p;
319
320	scratch = alloc_page(gfp_mask);
321	if (!scratch)
322		goto out;
323
324	xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
325	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
326
327	p = xdr_inline_decode(&xdr, sizeof(__be32));
328	if (!p)
329		goto out_free_scratch;
330	nr_volumes = be32_to_cpup(p++);
331
332	volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
333			  gfp_mask);
334	if (!volumes)
335		goto out_free_scratch;
336
337	for (i = 0; i < nr_volumes; i++) {
338		ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
339		if (ret < 0)
340			goto out_free_volumes;
341	}
342
343	top = kzalloc(sizeof(*top), gfp_mask);
344	if (!top)
345		goto out_free_volumes;
346
347	ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
348	if (ret) {
349		bl_free_device(top);
350		kfree(top);
351		goto out_free_volumes;
352	}
353
354	node = &top->node;
355	nfs4_init_deviceid_node(node, server, &pdev->dev_id);
356
357out_free_volumes:
358	kfree(volumes);
359out_free_scratch:
360	__free_page(scratch);
361out:
362	return node;
363}
364