• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * NVDIMM Block Window Driver
4  * Copyright (c) 2014, Intel Corporation.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/fs.h>
9 #include <linux/genhd.h>
10 #include <linux/module.h>
11 #include <linux/moduleparam.h>
12 #include <linux/nd.h>
13 #include <linux/sizes.h>
14 #include "nd.h"
15 
nsblk_meta_size(struct nd_namespace_blk * nsblk)16 static u32 nsblk_meta_size(struct nd_namespace_blk *nsblk)
17 {
18 	return nsblk->lbasize - ((nsblk->lbasize >= 4096) ? 4096 : 512);
19 }
20 
nsblk_internal_lbasize(struct nd_namespace_blk * nsblk)21 static u32 nsblk_internal_lbasize(struct nd_namespace_blk *nsblk)
22 {
23 	return roundup(nsblk->lbasize, INT_LBASIZE_ALIGNMENT);
24 }
25 
nsblk_sector_size(struct nd_namespace_blk * nsblk)26 static u32 nsblk_sector_size(struct nd_namespace_blk *nsblk)
27 {
28 	return nsblk->lbasize - nsblk_meta_size(nsblk);
29 }
30 
to_dev_offset(struct nd_namespace_blk * nsblk,resource_size_t ns_offset,unsigned int len)31 static resource_size_t to_dev_offset(struct nd_namespace_blk *nsblk,
32 				resource_size_t ns_offset, unsigned int len)
33 {
34 	int i;
35 
36 	for (i = 0; i < nsblk->num_resources; i++) {
37 		if (ns_offset < resource_size(nsblk->res[i])) {
38 			if (ns_offset + len > resource_size(nsblk->res[i])) {
39 				dev_WARN_ONCE(&nsblk->common.dev, 1,
40 					"illegal request\n");
41 				return SIZE_MAX;
42 			}
43 			return nsblk->res[i]->start + ns_offset;
44 		}
45 		ns_offset -= resource_size(nsblk->res[i]);
46 	}
47 
48 	dev_WARN_ONCE(&nsblk->common.dev, 1, "request out of range\n");
49 	return SIZE_MAX;
50 }
51 
to_ndbr(struct nd_namespace_blk * nsblk)52 static struct nd_blk_region *to_ndbr(struct nd_namespace_blk *nsblk)
53 {
54 	struct nd_region *nd_region;
55 	struct device *parent;
56 
57 	parent = nsblk->common.dev.parent;
58 	nd_region = container_of(parent, struct nd_region, dev);
59 	return container_of(nd_region, struct nd_blk_region, nd_region);
60 }
61 
62 #ifdef CONFIG_BLK_DEV_INTEGRITY
nd_blk_rw_integrity(struct nd_namespace_blk * nsblk,struct bio_integrity_payload * bip,u64 lba,int rw)63 static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
64 		struct bio_integrity_payload *bip, u64 lba, int rw)
65 {
66 	struct nd_blk_region *ndbr = to_ndbr(nsblk);
67 	unsigned int len = nsblk_meta_size(nsblk);
68 	resource_size_t	dev_offset, ns_offset;
69 	u32 internal_lbasize, sector_size;
70 	int err = 0;
71 
72 	internal_lbasize = nsblk_internal_lbasize(nsblk);
73 	sector_size = nsblk_sector_size(nsblk);
74 	ns_offset = lba * internal_lbasize + sector_size;
75 	dev_offset = to_dev_offset(nsblk, ns_offset, len);
76 	if (dev_offset == SIZE_MAX)
77 		return -EIO;
78 
79 	while (len) {
80 		unsigned int cur_len;
81 		struct bio_vec bv;
82 		void *iobuf;
83 
84 		bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
85 		/*
86 		 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
87 		 * .bv_offset already adjusted for iter->bi_bvec_done, and we
88 		 * can use those directly
89 		 */
90 
91 		cur_len = min(len, bv.bv_len);
92 		iobuf = kmap_atomic(bv.bv_page);
93 		err = ndbr->do_io(ndbr, dev_offset, iobuf + bv.bv_offset,
94 				cur_len, rw);
95 		kunmap_atomic(iobuf);
96 		if (err)
97 			return err;
98 
99 		len -= cur_len;
100 		dev_offset += cur_len;
101 		if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
102 			return -EIO;
103 	}
104 
105 	return err;
106 }
107 
108 #else /* CONFIG_BLK_DEV_INTEGRITY */
nd_blk_rw_integrity(struct nd_namespace_blk * nsblk,struct bio_integrity_payload * bip,u64 lba,int rw)109 static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
110 		struct bio_integrity_payload *bip, u64 lba, int rw)
111 {
112 	return 0;
113 }
114 #endif
115 
nsblk_do_bvec(struct nd_namespace_blk * nsblk,struct bio_integrity_payload * bip,struct page * page,unsigned int len,unsigned int off,int rw,sector_t sector)116 static int nsblk_do_bvec(struct nd_namespace_blk *nsblk,
117 		struct bio_integrity_payload *bip, struct page *page,
118 		unsigned int len, unsigned int off, int rw, sector_t sector)
119 {
120 	struct nd_blk_region *ndbr = to_ndbr(nsblk);
121 	resource_size_t	dev_offset, ns_offset;
122 	u32 internal_lbasize, sector_size;
123 	int err = 0;
124 	void *iobuf;
125 	u64 lba;
126 
127 	internal_lbasize = nsblk_internal_lbasize(nsblk);
128 	sector_size = nsblk_sector_size(nsblk);
129 	while (len) {
130 		unsigned int cur_len;
131 
132 		/*
133 		 * If we don't have an integrity payload, we don't have to
134 		 * split the bvec into sectors, as this would cause unnecessary
135 		 * Block Window setup/move steps. the do_io routine is capable
136 		 * of handling len <= PAGE_SIZE.
137 		 */
138 		cur_len = bip ? min(len, sector_size) : len;
139 
140 		lba = div_u64(sector << SECTOR_SHIFT, sector_size);
141 		ns_offset = lba * internal_lbasize;
142 		dev_offset = to_dev_offset(nsblk, ns_offset, cur_len);
143 		if (dev_offset == SIZE_MAX)
144 			return -EIO;
145 
146 		iobuf = kmap_atomic(page);
147 		err = ndbr->do_io(ndbr, dev_offset, iobuf + off, cur_len, rw);
148 		kunmap_atomic(iobuf);
149 		if (err)
150 			return err;
151 
152 		if (bip) {
153 			err = nd_blk_rw_integrity(nsblk, bip, lba, rw);
154 			if (err)
155 				return err;
156 		}
157 		len -= cur_len;
158 		off += cur_len;
159 		sector += sector_size >> SECTOR_SHIFT;
160 	}
161 
162 	return err;
163 }
164 
nd_blk_make_request(struct request_queue * q,struct bio * bio)165 static blk_qc_t nd_blk_make_request(struct request_queue *q, struct bio *bio)
166 {
167 	struct bio_integrity_payload *bip;
168 	struct nd_namespace_blk *nsblk;
169 	struct bvec_iter iter;
170 	unsigned long start;
171 	struct bio_vec bvec;
172 	int err = 0, rw;
173 	bool do_acct;
174 
175 	if (!bio_integrity_prep(bio))
176 		return BLK_QC_T_NONE;
177 
178 	bip = bio_integrity(bio);
179 	nsblk = q->queuedata;
180 	rw = bio_data_dir(bio);
181 	do_acct = nd_iostat_start(bio, &start);
182 	bio_for_each_segment(bvec, bio, iter) {
183 		unsigned int len = bvec.bv_len;
184 
185 		BUG_ON(len > PAGE_SIZE);
186 		err = nsblk_do_bvec(nsblk, bip, bvec.bv_page, len,
187 				bvec.bv_offset, rw, iter.bi_sector);
188 		if (err) {
189 			dev_dbg(&nsblk->common.dev,
190 					"io error in %s sector %lld, len %d,\n",
191 					(rw == READ) ? "READ" : "WRITE",
192 					(unsigned long long) iter.bi_sector, len);
193 			bio->bi_status = errno_to_blk_status(err);
194 			break;
195 		}
196 	}
197 	if (do_acct)
198 		nd_iostat_end(bio, start);
199 
200 	bio_endio(bio);
201 	return BLK_QC_T_NONE;
202 }
203 
nsblk_rw_bytes(struct nd_namespace_common * ndns,resource_size_t offset,void * iobuf,size_t n,int rw,unsigned long flags)204 static int nsblk_rw_bytes(struct nd_namespace_common *ndns,
205 		resource_size_t offset, void *iobuf, size_t n, int rw,
206 		unsigned long flags)
207 {
208 	struct nd_namespace_blk *nsblk = to_nd_namespace_blk(&ndns->dev);
209 	struct nd_blk_region *ndbr = to_ndbr(nsblk);
210 	resource_size_t	dev_offset;
211 
212 	dev_offset = to_dev_offset(nsblk, offset, n);
213 
214 	if (unlikely(offset + n > nsblk->size)) {
215 		dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
216 		return -EFAULT;
217 	}
218 
219 	if (dev_offset == SIZE_MAX)
220 		return -EIO;
221 
222 	return ndbr->do_io(ndbr, dev_offset, iobuf, n, rw);
223 }
224 
225 static const struct block_device_operations nd_blk_fops = {
226 	.owner = THIS_MODULE,
227 	.revalidate_disk = nvdimm_revalidate_disk,
228 };
229 
nd_blk_release_queue(void * q)230 static void nd_blk_release_queue(void *q)
231 {
232 	blk_cleanup_queue(q);
233 }
234 
nd_blk_release_disk(void * disk)235 static void nd_blk_release_disk(void *disk)
236 {
237 	del_gendisk(disk);
238 	put_disk(disk);
239 }
240 
nsblk_attach_disk(struct nd_namespace_blk * nsblk)241 static int nsblk_attach_disk(struct nd_namespace_blk *nsblk)
242 {
243 	struct device *dev = &nsblk->common.dev;
244 	resource_size_t available_disk_size;
245 	struct request_queue *q;
246 	struct gendisk *disk;
247 	u64 internal_nlba;
248 
249 	internal_nlba = div_u64(nsblk->size, nsblk_internal_lbasize(nsblk));
250 	available_disk_size = internal_nlba * nsblk_sector_size(nsblk);
251 
252 	q = blk_alloc_queue(GFP_KERNEL);
253 	if (!q)
254 		return -ENOMEM;
255 	if (devm_add_action_or_reset(dev, nd_blk_release_queue, q))
256 		return -ENOMEM;
257 
258 	blk_queue_make_request(q, nd_blk_make_request);
259 	blk_queue_max_hw_sectors(q, UINT_MAX);
260 	blk_queue_logical_block_size(q, nsblk_sector_size(nsblk));
261 	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
262 	q->queuedata = nsblk;
263 
264 	disk = alloc_disk(0);
265 	if (!disk)
266 		return -ENOMEM;
267 
268 	disk->first_minor	= 0;
269 	disk->fops		= &nd_blk_fops;
270 	disk->queue		= q;
271 	disk->flags		= GENHD_FL_EXT_DEVT;
272 	nvdimm_namespace_disk_name(&nsblk->common, disk->disk_name);
273 
274 	if (devm_add_action_or_reset(dev, nd_blk_release_disk, disk))
275 		return -ENOMEM;
276 
277 	if (nsblk_meta_size(nsblk)) {
278 		int rc = nd_integrity_init(disk, nsblk_meta_size(nsblk));
279 
280 		if (rc)
281 			return rc;
282 	}
283 
284 	set_capacity(disk, available_disk_size >> SECTOR_SHIFT);
285 	device_add_disk(dev, disk, NULL);
286 	revalidate_disk(disk);
287 	return 0;
288 }
289 
nd_blk_probe(struct device * dev)290 static int nd_blk_probe(struct device *dev)
291 {
292 	struct nd_namespace_common *ndns;
293 	struct nd_namespace_blk *nsblk;
294 
295 	ndns = nvdimm_namespace_common_probe(dev);
296 	if (IS_ERR(ndns))
297 		return PTR_ERR(ndns);
298 
299 	nsblk = to_nd_namespace_blk(&ndns->dev);
300 	nsblk->size = nvdimm_namespace_capacity(ndns);
301 	dev_set_drvdata(dev, nsblk);
302 
303 	ndns->rw_bytes = nsblk_rw_bytes;
304 	if (is_nd_btt(dev))
305 		return nvdimm_namespace_attach_btt(ndns);
306 	else if (nd_btt_probe(dev, ndns) == 0) {
307 		/* we'll come back as btt-blk */
308 		return -ENXIO;
309 	} else
310 		return nsblk_attach_disk(nsblk);
311 }
312 
nd_blk_remove(struct device * dev)313 static int nd_blk_remove(struct device *dev)
314 {
315 	if (is_nd_btt(dev))
316 		nvdimm_namespace_detach_btt(to_nd_btt(dev));
317 	return 0;
318 }
319 
320 static struct nd_device_driver nd_blk_driver = {
321 	.probe = nd_blk_probe,
322 	.remove = nd_blk_remove,
323 	.drv = {
324 		.name = "nd_blk",
325 	},
326 	.type = ND_DRIVER_NAMESPACE_BLK,
327 };
328 
nd_blk_init(void)329 static int __init nd_blk_init(void)
330 {
331 	return nd_driver_register(&nd_blk_driver);
332 }
333 
nd_blk_exit(void)334 static void __exit nd_blk_exit(void)
335 {
336 	driver_unregister(&nd_blk_driver.drv);
337 }
338 
339 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
340 MODULE_LICENSE("GPL v2");
341 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_BLK);
342 module_init(nd_blk_init);
343 module_exit(nd_blk_exit);
344