xref: /openbmc/linux/fs/nfs/blocklayout/dev.c (revision 77a87824)
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 || b->simple.nr_sigs > PNFS_BLOCK_MAX_UUIDS) {
69 			dprintk("Bad signature count: %d\n", b->simple.nr_sigs);
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 + \
93 				(XDR_QUADLEN(b->simple.sigs[i].sig_len) << 2);
94 		}
95 		break;
96 	case PNFS_BLOCK_VOLUME_SLICE:
97 		p = xdr_inline_decode(xdr, 8 + 8 + 4);
98 		if (!p)
99 			return -EIO;
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++);
103 		break;
104 	case PNFS_BLOCK_VOLUME_CONCAT:
105 		p = xdr_inline_decode(xdr, 4);
106 		if (!p)
107 			return -EIO;
108 
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);
112 			return -EIO;
113 		}
114 
115 		p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
116 		if (!p)
117 			return -EIO;
118 		for (i = 0; i < b->concat.volumes_count; i++)
119 			b->concat.volumes[i] = be32_to_cpup(p++);
120 		break;
121 	case PNFS_BLOCK_VOLUME_STRIPE:
122 		p = xdr_inline_decode(xdr, 8 + 4);
123 		if (!p)
124 			return -EIO;
125 
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);
130 			return -EIO;
131 		}
132 
133 		p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
134 		if (!p)
135 			return -EIO;
136 		for (i = 0; i < b->stripe.volumes_count; i++)
137 			b->stripe.volumes[i] = be32_to_cpup(p++);
138 		break;
139 	case PNFS_BLOCK_VOLUME_SCSI:
140 		p = xdr_inline_decode(xdr, 4 + 4 + 4);
141 		if (!p)
142 			return -EIO;
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);
147 		if (!p)
148 			return -EIO;
149 		if (b->scsi.designator_len > 256)
150 			return -EIO;
151 		memcpy(&b->scsi.designator, p, b->scsi.designator_len);
152 		p = xdr_inline_decode(xdr, 8);
153 		if (!p)
154 			return -EIO;
155 		p = xdr_decode_hyper(p, &b->scsi.pr_key);
156 		break;
157 	default:
158 		dprintk("unknown volume type!\n");
159 		return -EIO;
160 	}
161 
162 	return 0;
163 }
164 
165 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
166 		struct pnfs_block_dev_map *map)
167 {
168 	map->start = dev->start;
169 	map->len = dev->len;
170 	map->disk_offset = dev->disk_offset;
171 	map->bdev = dev->bdev;
172 	return true;
173 }
174 
175 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
176 		struct pnfs_block_dev_map *map)
177 {
178 	int i;
179 
180 	for (i = 0; i < dev->nr_children; i++) {
181 		struct pnfs_block_dev *child = &dev->children[i];
182 
183 		if (child->start > offset ||
184 		    child->start + child->len <= offset)
185 			continue;
186 
187 		child->map(child, offset - child->start, map);
188 		return true;
189 	}
190 
191 	dprintk("%s: ran off loop!\n", __func__);
192 	return false;
193 }
194 
195 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
196 		struct pnfs_block_dev_map *map)
197 {
198 	struct pnfs_block_dev *child;
199 	u64 chunk;
200 	u32 chunk_idx;
201 	u64 disk_offset;
202 
203 	chunk = div_u64(offset, dev->chunk_size);
204 	div_u64_rem(chunk, dev->nr_children, &chunk_idx);
205 
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 */
210 		return false;
211 	}
212 
213 	/* truncate offset to the beginning of the stripe */
214 	offset = chunk * dev->chunk_size;
215 
216 	/* disk offset of the stripe */
217 	disk_offset = div_u64(offset, dev->nr_children);
218 
219 	child = &dev->children[chunk_idx];
220 	child->map(child, disk_offset, map);
221 
222 	map->start += offset;
223 	map->disk_offset += disk_offset;
224 	map->len = dev->chunk_size;
225 	return true;
226 }
227 
228 static int
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);
231 
232 
233 static int
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)
236 {
237 	struct pnfs_block_volume *v = &volumes[idx];
238 	struct block_device *bdev;
239 	dev_t dev;
240 
241 	dev = bl_resolve_deviceid(server, v, gfp_mask);
242 	if (!dev)
243 		return -EIO;
244 
245 	bdev = blkdev_get_by_dev(dev, FMODE_READ | FMODE_WRITE, NULL);
246 	if (IS_ERR(bdev)) {
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);
250 	}
251 	d->bdev = bdev;
252 
253 
254 	d->len = i_size_read(d->bdev->bd_inode);
255 	d->map = bl_map_simple;
256 
257 	printk(KERN_INFO "pNFS: using block device %s\n",
258 		d->bdev->bd_disk->disk_name);
259 	return 0;
260 }
261 
262 static bool
263 bl_validate_designator(struct pnfs_block_volume *v)
264 {
265 	switch (v->scsi.designator_type) {
266 	case PS_DESIGNATOR_EUI64:
267 		if (v->scsi.code_set != PS_CODE_SET_BINARY)
268 			return false;
269 
270 		if (v->scsi.designator_len != 8 &&
271 		    v->scsi.designator_len != 10 &&
272 		    v->scsi.designator_len != 16)
273 			return false;
274 
275 		return true;
276 	case PS_DESIGNATOR_NAA:
277 		if (v->scsi.code_set != PS_CODE_SET_BINARY)
278 			return false;
279 
280 		if (v->scsi.designator_len != 8 &&
281 		    v->scsi.designator_len != 16)
282 			return false;
283 
284 		return true;
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",
289 			v->scsi.code_set,
290 			v->scsi.designator_type,
291 			v->scsi.designator_len);
292 		return false;
293 	default:
294 		pr_err("pNFS: invalid designator "
295 			"(code set %d, type %d, len %d.\n",
296 			v->scsi.code_set,
297 			v->scsi.designator_type,
298 			v->scsi.designator_len);
299 		return false;
300 	}
301 }
302 
303 /*
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.
306  */
307 static struct block_device *
308 bl_open_udev_path(struct pnfs_block_volume *v)
309 {
310 	struct block_device *bdev;
311 	const char *devname;
312 
313 	devname = kasprintf(GFP_KERNEL, "/dev/disk/by-id/wwn-0x%*phN",
314 				v->scsi.designator_len, v->scsi.designator);
315 	if (!devname)
316 		return ERR_PTR(-ENOMEM);
317 
318 	bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
319 	if (IS_ERR(bdev)) {
320 		pr_warn("pNFS: failed to open device %s (%ld)\n",
321 			devname, PTR_ERR(bdev));
322 	}
323 
324 	kfree(devname);
325 	return bdev;
326 }
327 
328 /*
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.
331  */
332 static struct block_device *
333 bl_open_dm_mpath_udev_path(struct pnfs_block_volume *v)
334 {
335 	struct block_device *bdev;
336 	const char *devname;
337 
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);
342 	if (!devname)
343 		return ERR_PTR(-ENOMEM);
344 
345 	bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
346 	kfree(devname);
347 	return bdev;
348 }
349 
350 static int
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)
353 {
354 	struct pnfs_block_volume *v = &volumes[idx];
355 	struct block_device *bdev;
356 	const struct pr_ops *ops;
357 	int error;
358 
359 	if (!bl_validate_designator(v))
360 		return -EINVAL;
361 
362 	bdev = bl_open_dm_mpath_udev_path(v);
363 	if (IS_ERR(bdev))
364 		bdev = bl_open_udev_path(v);
365 	if (IS_ERR(bdev))
366 		return PTR_ERR(bdev);
367 	d->bdev = bdev;
368 
369 	d->len = i_size_read(d->bdev->bd_inode);
370 	d->map = bl_map_simple;
371 	d->pr_key = v->scsi.pr_key;
372 
373 	pr_info("pNFS: using block device %s (reservation key 0x%llx)\n",
374 		d->bdev->bd_disk->disk_name, d->pr_key);
375 
376 	ops = d->bdev->bd_disk->fops->pr_ops;
377 	if (!ops) {
378 		pr_err("pNFS: block device %s does not support reservations.",
379 				d->bdev->bd_disk->disk_name);
380 		error = -EINVAL;
381 		goto out_blkdev_put;
382 	}
383 
384 	error = ops->pr_register(d->bdev, 0, d->pr_key, true);
385 	if (error) {
386 		pr_err("pNFS: failed to register key for block device %s.",
387 				d->bdev->bd_disk->disk_name);
388 		goto out_blkdev_put;
389 	}
390 
391 	d->pr_registered = true;
392 	return 0;
393 
394 out_blkdev_put:
395 	blkdev_put(d->bdev, FMODE_READ | FMODE_WRITE);
396 	return error;
397 }
398 
399 static int
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)
402 {
403 	struct pnfs_block_volume *v = &volumes[idx];
404 	int ret;
405 
406 	ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
407 	if (ret)
408 		return ret;
409 
410 	d->disk_offset = v->slice.start;
411 	d->len = v->slice.len;
412 	return 0;
413 }
414 
415 static int
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)
418 {
419 	struct pnfs_block_volume *v = &volumes[idx];
420 	u64 len = 0;
421 	int ret, i;
422 
423 	d->children = kcalloc(v->concat.volumes_count,
424 			sizeof(struct pnfs_block_dev), GFP_KERNEL);
425 	if (!d->children)
426 		return -ENOMEM;
427 
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);
431 		if (ret)
432 			return ret;
433 
434 		d->nr_children++;
435 		d->children[i].start += len;
436 		len += d->children[i].len;
437 	}
438 
439 	d->len = len;
440 	d->map = bl_map_concat;
441 	return 0;
442 }
443 
444 static int
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)
447 {
448 	struct pnfs_block_volume *v = &volumes[idx];
449 	u64 len = 0;
450 	int ret, i;
451 
452 	d->children = kcalloc(v->stripe.volumes_count,
453 			sizeof(struct pnfs_block_dev), GFP_KERNEL);
454 	if (!d->children)
455 		return -ENOMEM;
456 
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);
460 		if (ret)
461 			return ret;
462 
463 		d->nr_children++;
464 		len += d->children[i].len;
465 	}
466 
467 	d->len = len;
468 	d->chunk_size = v->stripe.chunk_size;
469 	d->map = bl_map_stripe;
470 	return 0;
471 }
472 
473 static int
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)
476 {
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);
488 	default:
489 		dprintk("unsupported volume type: %d\n", volumes[idx].type);
490 		return -EIO;
491 	}
492 }
493 
494 struct nfs4_deviceid_node *
495 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
496 		gfp_t gfp_mask)
497 {
498 	struct nfs4_deviceid_node *node = NULL;
499 	struct pnfs_block_volume *volumes;
500 	struct pnfs_block_dev *top;
501 	struct xdr_stream xdr;
502 	struct xdr_buf buf;
503 	struct page *scratch;
504 	int nr_volumes, ret, i;
505 	__be32 *p;
506 
507 	scratch = alloc_page(gfp_mask);
508 	if (!scratch)
509 		goto out;
510 
511 	xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
512 	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
513 
514 	p = xdr_inline_decode(&xdr, sizeof(__be32));
515 	if (!p)
516 		goto out_free_scratch;
517 	nr_volumes = be32_to_cpup(p++);
518 
519 	volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
520 			  gfp_mask);
521 	if (!volumes)
522 		goto out_free_scratch;
523 
524 	for (i = 0; i < nr_volumes; i++) {
525 		ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
526 		if (ret < 0)
527 			goto out_free_volumes;
528 	}
529 
530 	top = kzalloc(sizeof(*top), gfp_mask);
531 	if (!top)
532 		goto out_free_volumes;
533 
534 	ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
535 	if (ret) {
536 		bl_free_device(top);
537 		kfree(top);
538 		goto out_free_volumes;
539 	}
540 
541 	node = &top->node;
542 	nfs4_init_deviceid_node(node, server, &pdev->dev_id);
543 
544 out_free_volumes:
545 	kfree(volumes);
546 out_free_scratch:
547 	__free_page(scratch);
548 out:
549 	return node;
550 }
551