xref: /linux/fs/nfs/blocklayout/dev.c (revision c22760885fd6f7161fabde8e7db63d7f085b125a) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2014-2016 Christoph Hellwig.
4  */
5 #include <linux/sunrpc/svc.h>
6 #include <linux/blkdev.h>
7 #include <linux/nfs4.h>
8 #include <linux/nfs_fs.h>
9 #include <linux/nfs_xdr.h>
10 #include <linux/pr.h>
11 
12 #include "blocklayout.h"
13 
14 #define NFSDBG_FACILITY		NFSDBG_PNFS_LD
15 
16 static void
17 bl_free_device(struct pnfs_block_dev *dev)
18 {
19 	if (dev->nr_children) {
20 		int i;
21 
22 		for (i = 0; i < dev->nr_children; i++)
23 			bl_free_device(&dev->children[i]);
24 		kfree(dev->children);
25 	} else {
26 		if (dev->pr_registered) {
27 			const struct pr_ops *ops =
28 				dev->bdev->bd_disk->fops->pr_ops;
29 			int error;
30 
31 			error = ops->pr_register(dev->bdev, dev->pr_key, 0,
32 				false);
33 			if (error)
34 				pr_err("failed to unregister PR key.\n");
35 		}
36 
37 		if (dev->bdev)
38 			blkdev_put(dev->bdev, FMODE_READ | FMODE_WRITE);
39 	}
40 }
41 
42 void
43 bl_free_deviceid_node(struct nfs4_deviceid_node *d)
44 {
45 	struct pnfs_block_dev *dev =
46 		container_of(d, struct pnfs_block_dev, node);
47 
48 	bl_free_device(dev);
49 	kfree_rcu(dev, node.rcu);
50 }
51 
52 static int
53 nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b)
54 {
55 	__be32 *p;
56 	int i;
57 
58 	p = xdr_inline_decode(xdr, 4);
59 	if (!p)
60 		return -EIO;
61 	b->type = be32_to_cpup(p++);
62 
63 	switch (b->type) {
64 	case PNFS_BLOCK_VOLUME_SIMPLE:
65 		p = xdr_inline_decode(xdr, 4);
66 		if (!p)
67 			return -EIO;
68 		b->simple.nr_sigs = be32_to_cpup(p++);
69 		if (!b->simple.nr_sigs || b->simple.nr_sigs > PNFS_BLOCK_MAX_UUIDS) {
70 			dprintk("Bad signature count: %d\n", b->simple.nr_sigs);
71 			return -EIO;
72 		}
73 
74 		b->simple.len = 4 + 4;
75 		for (i = 0; i < b->simple.nr_sigs; i++) {
76 			p = xdr_inline_decode(xdr, 8 + 4);
77 			if (!p)
78 				return -EIO;
79 			p = xdr_decode_hyper(p, &b->simple.sigs[i].offset);
80 			b->simple.sigs[i].sig_len = be32_to_cpup(p++);
81 			if (b->simple.sigs[i].sig_len > PNFS_BLOCK_UUID_LEN) {
82 				pr_info("signature too long: %d\n",
83 					b->simple.sigs[i].sig_len);
84 				return -EIO;
85 			}
86 
87 			p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
88 			if (!p)
89 				return -EIO;
90 			memcpy(&b->simple.sigs[i].sig, p,
91 				b->simple.sigs[i].sig_len);
92 
93 			b->simple.len += 8 + 4 + \
94 				(XDR_QUADLEN(b->simple.sigs[i].sig_len) << 2);
95 		}
96 		break;
97 	case PNFS_BLOCK_VOLUME_SLICE:
98 		p = xdr_inline_decode(xdr, 8 + 8 + 4);
99 		if (!p)
100 			return -EIO;
101 		p = xdr_decode_hyper(p, &b->slice.start);
102 		p = xdr_decode_hyper(p, &b->slice.len);
103 		b->slice.volume = be32_to_cpup(p++);
104 		break;
105 	case PNFS_BLOCK_VOLUME_CONCAT:
106 		p = xdr_inline_decode(xdr, 4);
107 		if (!p)
108 			return -EIO;
109 
110 		b->concat.volumes_count = be32_to_cpup(p++);
111 		if (b->concat.volumes_count > PNFS_BLOCK_MAX_DEVICES) {
112 			dprintk("Too many volumes: %d\n", b->concat.volumes_count);
113 			return -EIO;
114 		}
115 
116 		p = xdr_inline_decode(xdr, b->concat.volumes_count * 4);
117 		if (!p)
118 			return -EIO;
119 		for (i = 0; i < b->concat.volumes_count; i++)
120 			b->concat.volumes[i] = be32_to_cpup(p++);
121 		break;
122 	case PNFS_BLOCK_VOLUME_STRIPE:
123 		p = xdr_inline_decode(xdr, 8 + 4);
124 		if (!p)
125 			return -EIO;
126 
127 		p = xdr_decode_hyper(p, &b->stripe.chunk_size);
128 		b->stripe.volumes_count = be32_to_cpup(p++);
129 		if (b->stripe.volumes_count > PNFS_BLOCK_MAX_DEVICES) {
130 			dprintk("Too many volumes: %d\n", b->stripe.volumes_count);
131 			return -EIO;
132 		}
133 
134 		p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4);
135 		if (!p)
136 			return -EIO;
137 		for (i = 0; i < b->stripe.volumes_count; i++)
138 			b->stripe.volumes[i] = be32_to_cpup(p++);
139 		break;
140 	case PNFS_BLOCK_VOLUME_SCSI:
141 		p = xdr_inline_decode(xdr, 4 + 4 + 4);
142 		if (!p)
143 			return -EIO;
144 		b->scsi.code_set = be32_to_cpup(p++);
145 		b->scsi.designator_type = be32_to_cpup(p++);
146 		b->scsi.designator_len = be32_to_cpup(p++);
147 		p = xdr_inline_decode(xdr, b->scsi.designator_len);
148 		if (!p)
149 			return -EIO;
150 		if (b->scsi.designator_len > 256)
151 			return -EIO;
152 		memcpy(&b->scsi.designator, p, b->scsi.designator_len);
153 		p = xdr_inline_decode(xdr, 8);
154 		if (!p)
155 			return -EIO;
156 		p = xdr_decode_hyper(p, &b->scsi.pr_key);
157 		break;
158 	default:
159 		dprintk("unknown volume type!\n");
160 		return -EIO;
161 	}
162 
163 	return 0;
164 }
165 
166 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
167 		struct pnfs_block_dev_map *map)
168 {
169 	map->start = dev->start;
170 	map->len = dev->len;
171 	map->disk_offset = dev->disk_offset;
172 	map->bdev = dev->bdev;
173 	return true;
174 }
175 
176 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset,
177 		struct pnfs_block_dev_map *map)
178 {
179 	int i;
180 
181 	for (i = 0; i < dev->nr_children; i++) {
182 		struct pnfs_block_dev *child = &dev->children[i];
183 
184 		if (child->start > offset ||
185 		    child->start + child->len <= offset)
186 			continue;
187 
188 		child->map(child, offset - child->start, map);
189 		return true;
190 	}
191 
192 	dprintk("%s: ran off loop!\n", __func__);
193 	return false;
194 }
195 
196 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset,
197 		struct pnfs_block_dev_map *map)
198 {
199 	struct pnfs_block_dev *child;
200 	u64 chunk;
201 	u32 chunk_idx;
202 	u64 disk_offset;
203 
204 	chunk = div_u64(offset, dev->chunk_size);
205 	div_u64_rem(chunk, dev->nr_children, &chunk_idx);
206 
207 	if (chunk_idx >= dev->nr_children) {
208 		dprintk("%s: invalid chunk idx %d (%lld/%lld)\n",
209 			__func__, chunk_idx, offset, dev->chunk_size);
210 		/* error, should not happen */
211 		return false;
212 	}
213 
214 	/* truncate offset to the beginning of the stripe */
215 	offset = chunk * dev->chunk_size;
216 
217 	/* disk offset of the stripe */
218 	disk_offset = div_u64(offset, dev->nr_children);
219 
220 	child = &dev->children[chunk_idx];
221 	child->map(child, disk_offset, map);
222 
223 	map->start += offset;
224 	map->disk_offset += disk_offset;
225 	map->len = dev->chunk_size;
226 	return true;
227 }
228 
229 static int
230 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
231 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
232 
233 
234 static int
235 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
236 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
237 {
238 	struct pnfs_block_volume *v = &volumes[idx];
239 	struct block_device *bdev;
240 	dev_t dev;
241 
242 	dev = bl_resolve_deviceid(server, v, gfp_mask);
243 	if (!dev)
244 		return -EIO;
245 
246 	bdev = blkdev_get_by_dev(dev, FMODE_READ | FMODE_WRITE, NULL);
247 	if (IS_ERR(bdev)) {
248 		printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
249 			MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
250 		return PTR_ERR(bdev);
251 	}
252 	d->bdev = bdev;
253 
254 
255 	d->len = bdev_nr_bytes(d->bdev);
256 	d->map = bl_map_simple;
257 
258 	printk(KERN_INFO "pNFS: using block device %s\n",
259 		d->bdev->bd_disk->disk_name);
260 	return 0;
261 }
262 
263 static bool
264 bl_validate_designator(struct pnfs_block_volume *v)
265 {
266 	switch (v->scsi.designator_type) {
267 	case PS_DESIGNATOR_EUI64:
268 		if (v->scsi.code_set != PS_CODE_SET_BINARY)
269 			return false;
270 
271 		if (v->scsi.designator_len != 8 &&
272 		    v->scsi.designator_len != 10 &&
273 		    v->scsi.designator_len != 16)
274 			return false;
275 
276 		return true;
277 	case PS_DESIGNATOR_NAA:
278 		if (v->scsi.code_set != PS_CODE_SET_BINARY)
279 			return false;
280 
281 		if (v->scsi.designator_len != 8 &&
282 		    v->scsi.designator_len != 16)
283 			return false;
284 
285 		return true;
286 	case PS_DESIGNATOR_T10:
287 	case PS_DESIGNATOR_NAME:
288 		pr_err("pNFS: unsupported designator "
289 			"(code set %d, type %d, len %d.\n",
290 			v->scsi.code_set,
291 			v->scsi.designator_type,
292 			v->scsi.designator_len);
293 		return false;
294 	default:
295 		pr_err("pNFS: invalid designator "
296 			"(code set %d, type %d, len %d.\n",
297 			v->scsi.code_set,
298 			v->scsi.designator_type,
299 			v->scsi.designator_len);
300 		return false;
301 	}
302 }
303 
304 /*
305  * Try to open the udev path for the WWN.  At least on Debian the udev
306  * by-id path will always point to the dm-multipath device if one exists.
307  */
308 static struct block_device *
309 bl_open_udev_path(struct pnfs_block_volume *v)
310 {
311 	struct block_device *bdev;
312 	const char *devname;
313 
314 	devname = kasprintf(GFP_KERNEL, "/dev/disk/by-id/wwn-0x%*phN",
315 				v->scsi.designator_len, v->scsi.designator);
316 	if (!devname)
317 		return ERR_PTR(-ENOMEM);
318 
319 	bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
320 	if (IS_ERR(bdev)) {
321 		pr_warn("pNFS: failed to open device %s (%ld)\n",
322 			devname, PTR_ERR(bdev));
323 	}
324 
325 	kfree(devname);
326 	return bdev;
327 }
328 
329 /*
330  * Try to open the RH/Fedora specific dm-mpath udev path for this WWN, as the
331  * wwn- links will only point to the first discovered SCSI device there.
332  */
333 static struct block_device *
334 bl_open_dm_mpath_udev_path(struct pnfs_block_volume *v)
335 {
336 	struct block_device *bdev;
337 	const char *devname;
338 
339 	devname = kasprintf(GFP_KERNEL,
340 			"/dev/disk/by-id/dm-uuid-mpath-%d%*phN",
341 			v->scsi.designator_type,
342 			v->scsi.designator_len, v->scsi.designator);
343 	if (!devname)
344 		return ERR_PTR(-ENOMEM);
345 
346 	bdev = blkdev_get_by_path(devname, FMODE_READ | FMODE_WRITE, NULL);
347 	kfree(devname);
348 	return bdev;
349 }
350 
351 static int
352 bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d,
353 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
354 {
355 	struct pnfs_block_volume *v = &volumes[idx];
356 	struct block_device *bdev;
357 	const struct pr_ops *ops;
358 	int error;
359 
360 	if (!bl_validate_designator(v))
361 		return -EINVAL;
362 
363 	bdev = bl_open_dm_mpath_udev_path(v);
364 	if (IS_ERR(bdev))
365 		bdev = bl_open_udev_path(v);
366 	if (IS_ERR(bdev))
367 		return PTR_ERR(bdev);
368 	d->bdev = bdev;
369 
370 	d->len = bdev_nr_bytes(d->bdev);
371 	d->map = bl_map_simple;
372 	d->pr_key = v->scsi.pr_key;
373 
374 	pr_info("pNFS: using block device %s (reservation key 0x%llx)\n",
375 		d->bdev->bd_disk->disk_name, d->pr_key);
376 
377 	ops = d->bdev->bd_disk->fops->pr_ops;
378 	if (!ops) {
379 		pr_err("pNFS: block device %s does not support reservations.",
380 				d->bdev->bd_disk->disk_name);
381 		error = -EINVAL;
382 		goto out_blkdev_put;
383 	}
384 
385 	error = ops->pr_register(d->bdev, 0, d->pr_key, true);
386 	if (error) {
387 		pr_err("pNFS: failed to register key for block device %s.",
388 				d->bdev->bd_disk->disk_name);
389 		goto out_blkdev_put;
390 	}
391 
392 	d->pr_registered = true;
393 	return 0;
394 
395 out_blkdev_put:
396 	blkdev_put(d->bdev, FMODE_READ | FMODE_WRITE);
397 	return error;
398 }
399 
400 static int
401 bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d,
402 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
403 {
404 	struct pnfs_block_volume *v = &volumes[idx];
405 	int ret;
406 
407 	ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask);
408 	if (ret)
409 		return ret;
410 
411 	d->disk_offset = v->slice.start;
412 	d->len = v->slice.len;
413 	return 0;
414 }
415 
416 static int
417 bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d,
418 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
419 {
420 	struct pnfs_block_volume *v = &volumes[idx];
421 	u64 len = 0;
422 	int ret, i;
423 
424 	d->children = kcalloc(v->concat.volumes_count,
425 			sizeof(struct pnfs_block_dev), GFP_KERNEL);
426 	if (!d->children)
427 		return -ENOMEM;
428 
429 	for (i = 0; i < v->concat.volumes_count; i++) {
430 		ret = bl_parse_deviceid(server, &d->children[i],
431 				volumes, v->concat.volumes[i], gfp_mask);
432 		if (ret)
433 			return ret;
434 
435 		d->nr_children++;
436 		d->children[i].start += len;
437 		len += d->children[i].len;
438 	}
439 
440 	d->len = len;
441 	d->map = bl_map_concat;
442 	return 0;
443 }
444 
445 static int
446 bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d,
447 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
448 {
449 	struct pnfs_block_volume *v = &volumes[idx];
450 	u64 len = 0;
451 	int ret, i;
452 
453 	d->children = kcalloc(v->stripe.volumes_count,
454 			sizeof(struct pnfs_block_dev), GFP_KERNEL);
455 	if (!d->children)
456 		return -ENOMEM;
457 
458 	for (i = 0; i < v->stripe.volumes_count; i++) {
459 		ret = bl_parse_deviceid(server, &d->children[i],
460 				volumes, v->stripe.volumes[i], gfp_mask);
461 		if (ret)
462 			return ret;
463 
464 		d->nr_children++;
465 		len += d->children[i].len;
466 	}
467 
468 	d->len = len;
469 	d->chunk_size = v->stripe.chunk_size;
470 	d->map = bl_map_stripe;
471 	return 0;
472 }
473 
474 static int
475 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d,
476 		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
477 {
478 	switch (volumes[idx].type) {
479 	case PNFS_BLOCK_VOLUME_SIMPLE:
480 		return bl_parse_simple(server, d, volumes, idx, gfp_mask);
481 	case PNFS_BLOCK_VOLUME_SLICE:
482 		return bl_parse_slice(server, d, volumes, idx, gfp_mask);
483 	case PNFS_BLOCK_VOLUME_CONCAT:
484 		return bl_parse_concat(server, d, volumes, idx, gfp_mask);
485 	case PNFS_BLOCK_VOLUME_STRIPE:
486 		return bl_parse_stripe(server, d, volumes, idx, gfp_mask);
487 	case PNFS_BLOCK_VOLUME_SCSI:
488 		return bl_parse_scsi(server, d, volumes, idx, gfp_mask);
489 	default:
490 		dprintk("unsupported volume type: %d\n", volumes[idx].type);
491 		return -EIO;
492 	}
493 }
494 
495 struct nfs4_deviceid_node *
496 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
497 		gfp_t gfp_mask)
498 {
499 	struct nfs4_deviceid_node *node = NULL;
500 	struct pnfs_block_volume *volumes;
501 	struct pnfs_block_dev *top;
502 	struct xdr_stream xdr;
503 	struct xdr_buf buf;
504 	struct page *scratch;
505 	int nr_volumes, ret, i;
506 	__be32 *p;
507 
508 	scratch = alloc_page(gfp_mask);
509 	if (!scratch)
510 		goto out;
511 
512 	xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen);
513 	xdr_set_scratch_page(&xdr, scratch);
514 
515 	p = xdr_inline_decode(&xdr, sizeof(__be32));
516 	if (!p)
517 		goto out_free_scratch;
518 	nr_volumes = be32_to_cpup(p++);
519 
520 	volumes = kcalloc(nr_volumes, sizeof(struct pnfs_block_volume),
521 			  gfp_mask);
522 	if (!volumes)
523 		goto out_free_scratch;
524 
525 	for (i = 0; i < nr_volumes; i++) {
526 		ret = nfs4_block_decode_volume(&xdr, &volumes[i]);
527 		if (ret < 0)
528 			goto out_free_volumes;
529 	}
530 
531 	top = kzalloc(sizeof(*top), gfp_mask);
532 	if (!top)
533 		goto out_free_volumes;
534 
535 	ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask);
536 
537 	node = &top->node;
538 	nfs4_init_deviceid_node(node, server, &pdev->dev_id);
539 	if (ret)
540 		nfs4_mark_deviceid_unavailable(node);
541 
542 out_free_volumes:
543 	kfree(volumes);
544 out_free_scratch:
545 	__free_page(scratch);
546 out:
547 	return node;
548 }
549