1/*
2 * NVDIMM Block Window Driver
3 * Copyright (c) 2014, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14
15#include <linux/blkdev.h>
16#include <linux/fs.h>
17#include <linux/genhd.h>
18#include <linux/module.h>
19#include <linux/moduleparam.h>
20#include <linux/nd.h>
21#include <linux/sizes.h>
22#include "nd.h"
23
24static u32 nsblk_meta_size(struct nd_namespace_blk *nsblk)
25{
26 return nsblk->lbasize - ((nsblk->lbasize >= 4096) ? 4096 : 512);
27}
28
29static u32 nsblk_internal_lbasize(struct nd_namespace_blk *nsblk)
30{
31 return roundup(nsblk->lbasize, INT_LBASIZE_ALIGNMENT);
32}
33
34static u32 nsblk_sector_size(struct nd_namespace_blk *nsblk)
35{
36 return nsblk->lbasize - nsblk_meta_size(nsblk);
37}
38
39static resource_size_t to_dev_offset(struct nd_namespace_blk *nsblk,
40 resource_size_t ns_offset, unsigned int len)
41{
42 int i;
43
44 for (i = 0; i < nsblk->num_resources; i++) {
45 if (ns_offset < resource_size(nsblk->res[i])) {
46 if (ns_offset + len > resource_size(nsblk->res[i])) {
47 dev_WARN_ONCE(&nsblk->common.dev, 1,
48 "illegal request\n");
49 return SIZE_MAX;
50 }
51 return nsblk->res[i]->start + ns_offset;
52 }
53 ns_offset -= resource_size(nsblk->res[i]);
54 }
55
56 dev_WARN_ONCE(&nsblk->common.dev, 1, "request out of range\n");
57 return SIZE_MAX;
58}
59
60static struct nd_blk_region *to_ndbr(struct nd_namespace_blk *nsblk)
61{
62 struct nd_region *nd_region;
63 struct device *parent;
64
65 parent = nsblk->common.dev.parent;
66 nd_region = container_of(parent, struct nd_region, dev);
67 return container_of(nd_region, struct nd_blk_region, nd_region);
68}
69
70#ifdef CONFIG_BLK_DEV_INTEGRITY
71static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
72 struct bio_integrity_payload *bip, u64 lba, int rw)
73{
74 struct nd_blk_region *ndbr = to_ndbr(nsblk);
75 unsigned int len = nsblk_meta_size(nsblk);
76 resource_size_t dev_offset, ns_offset;
77 u32 internal_lbasize, sector_size;
78 int err = 0;
79
80 internal_lbasize = nsblk_internal_lbasize(nsblk);
81 sector_size = nsblk_sector_size(nsblk);
82 ns_offset = lba * internal_lbasize + sector_size;
83 dev_offset = to_dev_offset(nsblk, ns_offset, len);
84 if (dev_offset == SIZE_MAX)
85 return -EIO;
86
87 while (len) {
88 unsigned int cur_len;
89 struct bio_vec bv;
90 void *iobuf;
91
92 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
93 /*
94 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
95 * .bv_offset already adjusted for iter->bi_bvec_done, and we
96 * can use those directly
97 */
98
99 cur_len = min(len, bv.bv_len);
100 iobuf = kmap_atomic(bv.bv_page);
101 err = ndbr->do_io(ndbr, dev_offset, iobuf + bv.bv_offset,
102 cur_len, rw);
103 kunmap_atomic(iobuf);
104 if (err)
105 return err;
106
107 len -= cur_len;
108 dev_offset += cur_len;
109 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
110 return -EIO;
111 }
112
113 return err;
114}
115
116#else /* CONFIG_BLK_DEV_INTEGRITY */
117static int nd_blk_rw_integrity(struct nd_namespace_blk *nsblk,
118 struct bio_integrity_payload *bip, u64 lba, int rw)
119{
120 return 0;
121}
122#endif
123
124static int nsblk_do_bvec(struct nd_namespace_blk *nsblk,
125 struct bio_integrity_payload *bip, struct page *page,
126 unsigned int len, unsigned int off, int rw, sector_t sector)
127{
128 struct nd_blk_region *ndbr = to_ndbr(nsblk);
129 resource_size_t dev_offset, ns_offset;
130 u32 internal_lbasize, sector_size;
131 int err = 0;
132 void *iobuf;
133 u64 lba;
134
135 internal_lbasize = nsblk_internal_lbasize(nsblk);
136 sector_size = nsblk_sector_size(nsblk);
137 while (len) {
138 unsigned int cur_len;
139
140 /*
141 * If we don't have an integrity payload, we don't have to
142 * split the bvec into sectors, as this would cause unnecessary
143 * Block Window setup/move steps. the do_io routine is capable
144 * of handling len <= PAGE_SIZE.
145 */
146 cur_len = bip ? min(len, sector_size) : len;
147
148 lba = div_u64(sector << SECTOR_SHIFT, sector_size);
149 ns_offset = lba * internal_lbasize;
150 dev_offset = to_dev_offset(nsblk, ns_offset, cur_len);
151 if (dev_offset == SIZE_MAX)
152 return -EIO;
153
154 iobuf = kmap_atomic(page);
155 err = ndbr->do_io(ndbr, dev_offset, iobuf + off, cur_len, rw);
156 kunmap_atomic(iobuf);
157 if (err)
158 return err;
159
160 if (bip) {
161 err = nd_blk_rw_integrity(nsblk, bip, lba, rw);
162 if (err)
163 return err;
164 }
165 len -= cur_len;
166 off += cur_len;
167 sector += sector_size >> SECTOR_SHIFT;
168 }
169
170 return err;
171}
172
173static blk_qc_t nd_blk_make_request(struct request_queue *q, struct bio *bio)
174{
175 struct bio_integrity_payload *bip;
176 struct nd_namespace_blk *nsblk;
177 struct bvec_iter iter;
178 unsigned long start;
179 struct bio_vec bvec;
180 int err = 0, rw;
181 bool do_acct;
182
183 if (!bio_integrity_prep(bio))
184 return BLK_QC_T_NONE;
185
186 bip = bio_integrity(bio);
187 nsblk = q->queuedata;
188 rw = bio_data_dir(bio);
189 do_acct = nd_iostat_start(bio, &start);
190 bio_for_each_segment(bvec, bio, iter) {
191 unsigned int len = bvec.bv_len;
192
193 BUG_ON(len > PAGE_SIZE);
194 err = nsblk_do_bvec(nsblk, bip, bvec.bv_page, len,
195 bvec.bv_offset, rw, iter.bi_sector);
196 if (err) {
197 dev_dbg(&nsblk->common.dev,
198 "io error in %s sector %lld, len %d,\n",
199 (rw == READ) ? "READ" : "WRITE",
200 (unsigned long long) iter.bi_sector, len);
201 bio->bi_status = errno_to_blk_status(err);
202 break;
203 }
204 }
205 if (do_acct)
206 nd_iostat_end(bio, start);
207
208 bio_endio(bio);
209 return BLK_QC_T_NONE;
210}
211
212static int nsblk_rw_bytes(struct nd_namespace_common *ndns,
213 resource_size_t offset, void *iobuf, size_t n, int rw,
214 unsigned long flags)
215{
216 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(&ndns->dev);
217 struct nd_blk_region *ndbr = to_ndbr(nsblk);
218 resource_size_t dev_offset;
219
220 dev_offset = to_dev_offset(nsblk, offset, n);
221
222 if (unlikely(offset + n > nsblk->size)) {
223 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
224 return -EFAULT;
225 }
226
227 if (dev_offset == SIZE_MAX)
228 return -EIO;
229
230 return ndbr->do_io(ndbr, dev_offset, iobuf, n, rw);
231}
232
233static const struct block_device_operations nd_blk_fops = {
234 .owner = THIS_MODULE,
235 .revalidate_disk = nvdimm_revalidate_disk,
236};
237
238static void nd_blk_release_queue(void *q)
239{
240 blk_cleanup_queue(q);
241}
242
243static void nd_blk_release_disk(void *disk)
244{
245 del_gendisk(disk);
246 put_disk(disk);
247}
248
249static int nsblk_attach_disk(struct nd_namespace_blk *nsblk)
250{
251 struct device *dev = &nsblk->common.dev;
252 resource_size_t available_disk_size;
253 struct request_queue *q;
254 struct gendisk *disk;
255 u64 internal_nlba;
256
257 internal_nlba = div_u64(nsblk->size, nsblk_internal_lbasize(nsblk));
258 available_disk_size = internal_nlba * nsblk_sector_size(nsblk);
259
260 q = blk_alloc_queue(GFP_KERNEL);
261 if (!q)
262 return -ENOMEM;
263 if (devm_add_action_or_reset(dev, nd_blk_release_queue, q))
264 return -ENOMEM;
265
266 blk_queue_make_request(q, nd_blk_make_request);
267 blk_queue_max_hw_sectors(q, UINT_MAX);
268 blk_queue_logical_block_size(q, nsblk_sector_size(nsblk));
269 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
270 q->queuedata = nsblk;
271
272 disk = alloc_disk(0);
273 if (!disk)
274 return -ENOMEM;
275
276 disk->first_minor = 0;
277 disk->fops = &nd_blk_fops;
278 disk->queue = q;
279 disk->flags = GENHD_FL_EXT_DEVT;
280 nvdimm_namespace_disk_name(&nsblk->common, disk->disk_name);
281
282 if (devm_add_action_or_reset(dev, nd_blk_release_disk, disk))
283 return -ENOMEM;
284
285 if (nsblk_meta_size(nsblk)) {
286 int rc = nd_integrity_init(disk, nsblk_meta_size(nsblk));
287
288 if (rc)
289 return rc;
290 }
291
292 set_capacity(disk, available_disk_size >> SECTOR_SHIFT);
293 device_add_disk(dev, disk, NULL);
294 revalidate_disk(disk);
295 return 0;
296}
297
298static int nd_blk_probe(struct device *dev)
299{
300 struct nd_namespace_common *ndns;
301 struct nd_namespace_blk *nsblk;
302
303 ndns = nvdimm_namespace_common_probe(dev);
304 if (IS_ERR(ndns))
305 return PTR_ERR(ndns);
306
307 nsblk = to_nd_namespace_blk(&ndns->dev);
308 nsblk->size = nvdimm_namespace_capacity(ndns);
309 dev_set_drvdata(dev, nsblk);
310
311 ndns->rw_bytes = nsblk_rw_bytes;
312 if (is_nd_btt(dev))
313 return nvdimm_namespace_attach_btt(ndns);
314 else if (nd_btt_probe(dev, ndns) == 0) {
315 /* we'll come back as btt-blk */
316 return -ENXIO;
317 } else
318 return nsblk_attach_disk(nsblk);
319}
320
321static int nd_blk_remove(struct device *dev)
322{
323 if (is_nd_btt(dev))
324 nvdimm_namespace_detach_btt(to_nd_btt(dev));
325 return 0;
326}
327
328static struct nd_device_driver nd_blk_driver = {
329 .probe = nd_blk_probe,
330 .remove = nd_blk_remove,
331 .drv = {
332 .name = "nd_blk",
333 },
334 .type = ND_DRIVER_NAMESPACE_BLK,
335};
336
337static int __init nd_blk_init(void)
338{
339 return nd_driver_register(&nd_blk_driver);
340}
341
342static void __exit nd_blk_exit(void)
343{
344 driver_unregister(&nd_blk_driver.drv);
345}
346
347MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
348MODULE_LICENSE("GPL v2");
349MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_BLK);
350module_init(nd_blk_init);
351module_exit(nd_blk_exit);
352