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
24 struct nd_blk_device {
25 struct request_queue *queue;
26 struct gendisk *disk;
27 struct nd_namespace_blk *nsblk;
28 struct nd_blk_region *ndbr;
29 size_t disk_size;
30 u32 sector_size;
31 u32 internal_lbasize;
32 };
33
34 static int nd_blk_major;
35
nd_blk_meta_size(struct nd_blk_device * blk_dev)36 static u32 nd_blk_meta_size(struct nd_blk_device *blk_dev)
37 {
38 return blk_dev->nsblk->lbasize - blk_dev->sector_size;
39 }
40
to_dev_offset(struct nd_namespace_blk * nsblk,resource_size_t ns_offset,unsigned int len)41 static resource_size_t to_dev_offset(struct nd_namespace_blk *nsblk,
42 resource_size_t ns_offset, unsigned int len)
43 {
44 int i;
45
46 for (i = 0; i < nsblk->num_resources; i++) {
47 if (ns_offset < resource_size(nsblk->res[i])) {
48 if (ns_offset + len > resource_size(nsblk->res[i])) {
49 dev_WARN_ONCE(&nsblk->common.dev, 1,
50 "illegal request\n");
51 return SIZE_MAX;
52 }
53 return nsblk->res[i]->start + ns_offset;
54 }
55 ns_offset -= resource_size(nsblk->res[i]);
56 }
57
58 dev_WARN_ONCE(&nsblk->common.dev, 1, "request out of range\n");
59 return SIZE_MAX;
60 }
61
62 #ifdef CONFIG_BLK_DEV_INTEGRITY
nd_blk_rw_integrity(struct nd_blk_device * blk_dev,struct bio_integrity_payload * bip,u64 lba,int rw)63 static int nd_blk_rw_integrity(struct nd_blk_device *blk_dev,
64 struct bio_integrity_payload *bip, u64 lba,
65 int rw)
66 {
67 unsigned int len = nd_blk_meta_size(blk_dev);
68 resource_size_t dev_offset, ns_offset;
69 struct nd_namespace_blk *nsblk;
70 struct nd_blk_region *ndbr;
71 int err = 0;
72
73 nsblk = blk_dev->nsblk;
74 ndbr = blk_dev->ndbr;
75 ns_offset = lba * blk_dev->internal_lbasize + blk_dev->sector_size;
76 dev_offset = to_dev_offset(nsblk, ns_offset, len);
77 if (dev_offset == SIZE_MAX)
78 return -EIO;
79
80 while (len) {
81 unsigned int cur_len;
82 struct bio_vec bv;
83 void *iobuf;
84
85 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
86 /*
87 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
88 * .bv_offset already adjusted for iter->bi_bvec_done, and we
89 * can use those directly
90 */
91
92 cur_len = min(len, bv.bv_len);
93 iobuf = kmap_atomic(bv.bv_page);
94 err = ndbr->do_io(ndbr, dev_offset, iobuf + bv.bv_offset,
95 cur_len, rw);
96 kunmap_atomic(iobuf);
97 if (err)
98 return err;
99
100 len -= cur_len;
101 dev_offset += cur_len;
102 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len);
103 }
104
105 return err;
106 }
107
108 #else /* CONFIG_BLK_DEV_INTEGRITY */
nd_blk_rw_integrity(struct nd_blk_device * blk_dev,struct bio_integrity_payload * bip,u64 lba,int rw)109 static int nd_blk_rw_integrity(struct nd_blk_device *blk_dev,
110 struct bio_integrity_payload *bip, u64 lba,
111 int rw)
112 {
113 return 0;
114 }
115 #endif
116
nd_blk_do_bvec(struct nd_blk_device * blk_dev,struct bio_integrity_payload * bip,struct page * page,unsigned int len,unsigned int off,int rw,sector_t sector)117 static int nd_blk_do_bvec(struct nd_blk_device *blk_dev,
118 struct bio_integrity_payload *bip, struct page *page,
119 unsigned int len, unsigned int off, int rw,
120 sector_t sector)
121 {
122 struct nd_blk_region *ndbr = blk_dev->ndbr;
123 resource_size_t dev_offset, ns_offset;
124 int err = 0;
125 void *iobuf;
126 u64 lba;
127
128 while (len) {
129 unsigned int cur_len;
130
131 /*
132 * If we don't have an integrity payload, we don't have to
133 * split the bvec into sectors, as this would cause unnecessary
134 * Block Window setup/move steps. the do_io routine is capable
135 * of handling len <= PAGE_SIZE.
136 */
137 cur_len = bip ? min(len, blk_dev->sector_size) : len;
138
139 lba = div_u64(sector << SECTOR_SHIFT, blk_dev->sector_size);
140 ns_offset = lba * blk_dev->internal_lbasize;
141 dev_offset = to_dev_offset(blk_dev->nsblk, ns_offset, cur_len);
142 if (dev_offset == SIZE_MAX)
143 return -EIO;
144
145 iobuf = kmap_atomic(page);
146 err = ndbr->do_io(ndbr, dev_offset, iobuf + off, cur_len, rw);
147 kunmap_atomic(iobuf);
148 if (err)
149 return err;
150
151 if (bip) {
152 err = nd_blk_rw_integrity(blk_dev, bip, lba, rw);
153 if (err)
154 return err;
155 }
156 len -= cur_len;
157 off += cur_len;
158 sector += blk_dev->sector_size >> SECTOR_SHIFT;
159 }
160
161 return err;
162 }
163
nd_blk_make_request(struct request_queue * q,struct bio * bio)164 static blk_qc_t nd_blk_make_request(struct request_queue *q, struct bio *bio)
165 {
166 struct block_device *bdev = bio->bi_bdev;
167 struct gendisk *disk = bdev->bd_disk;
168 struct bio_integrity_payload *bip;
169 struct nd_blk_device *blk_dev;
170 struct bvec_iter iter;
171 unsigned long start;
172 struct bio_vec bvec;
173 int err = 0, rw;
174 bool do_acct;
175
176 /*
177 * bio_integrity_enabled also checks if the bio already has an
178 * integrity payload attached. If it does, we *don't* do a
179 * bio_integrity_prep here - the payload has been generated by
180 * another kernel subsystem, and we just pass it through.
181 */
182 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
183 bio->bi_error = -EIO;
184 goto out;
185 }
186
187 bip = bio_integrity(bio);
188 blk_dev = disk->private_data;
189 rw = bio_data_dir(bio);
190 do_acct = nd_iostat_start(bio, &start);
191 bio_for_each_segment(bvec, bio, iter) {
192 unsigned int len = bvec.bv_len;
193
194 BUG_ON(len > PAGE_SIZE);
195 err = nd_blk_do_bvec(blk_dev, bip, bvec.bv_page, len,
196 bvec.bv_offset, rw, iter.bi_sector);
197 if (err) {
198 dev_info(&blk_dev->nsblk->common.dev,
199 "io error in %s sector %lld, len %d,\n",
200 (rw == READ) ? "READ" : "WRITE",
201 (unsigned long long) iter.bi_sector, len);
202 bio->bi_error = err;
203 break;
204 }
205 }
206 if (do_acct)
207 nd_iostat_end(bio, start);
208
209 out:
210 bio_endio(bio);
211 return BLK_QC_T_NONE;
212 }
213
nd_blk_rw_bytes(struct nd_namespace_common * ndns,resource_size_t offset,void * iobuf,size_t n,int rw)214 static int nd_blk_rw_bytes(struct nd_namespace_common *ndns,
215 resource_size_t offset, void *iobuf, size_t n, int rw)
216 {
217 struct nd_blk_device *blk_dev = dev_get_drvdata(ndns->claim);
218 struct nd_namespace_blk *nsblk = blk_dev->nsblk;
219 struct nd_blk_region *ndbr = blk_dev->ndbr;
220 resource_size_t dev_offset;
221
222 dev_offset = to_dev_offset(nsblk, offset, n);
223
224 if (unlikely(offset + n > blk_dev->disk_size)) {
225 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
226 return -EFAULT;
227 }
228
229 if (dev_offset == SIZE_MAX)
230 return -EIO;
231
232 return ndbr->do_io(ndbr, dev_offset, iobuf, n, rw);
233 }
234
235 static const struct block_device_operations nd_blk_fops = {
236 .owner = THIS_MODULE,
237 .revalidate_disk = nvdimm_revalidate_disk,
238 };
239
nd_blk_attach_disk(struct nd_namespace_common * ndns,struct nd_blk_device * blk_dev)240 static int nd_blk_attach_disk(struct nd_namespace_common *ndns,
241 struct nd_blk_device *blk_dev)
242 {
243 resource_size_t available_disk_size;
244 struct gendisk *disk;
245 u64 internal_nlba;
246
247 internal_nlba = div_u64(blk_dev->disk_size, blk_dev->internal_lbasize);
248 available_disk_size = internal_nlba * blk_dev->sector_size;
249
250 blk_dev->queue = blk_alloc_queue(GFP_KERNEL);
251 if (!blk_dev->queue)
252 return -ENOMEM;
253
254 blk_queue_make_request(blk_dev->queue, nd_blk_make_request);
255 blk_queue_max_hw_sectors(blk_dev->queue, UINT_MAX);
256 blk_queue_bounce_limit(blk_dev->queue, BLK_BOUNCE_ANY);
257 blk_queue_logical_block_size(blk_dev->queue, blk_dev->sector_size);
258 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, blk_dev->queue);
259
260 disk = blk_dev->disk = alloc_disk(0);
261 if (!disk) {
262 blk_cleanup_queue(blk_dev->queue);
263 return -ENOMEM;
264 }
265
266 disk->driverfs_dev = &ndns->dev;
267 disk->major = nd_blk_major;
268 disk->first_minor = 0;
269 disk->fops = &nd_blk_fops;
270 disk->private_data = blk_dev;
271 disk->queue = blk_dev->queue;
272 disk->flags = GENHD_FL_EXT_DEVT;
273 nvdimm_namespace_disk_name(ndns, disk->disk_name);
274 set_capacity(disk, 0);
275 add_disk(disk);
276
277 if (nd_blk_meta_size(blk_dev)) {
278 int rc = nd_integrity_init(disk, nd_blk_meta_size(blk_dev));
279
280 if (rc) {
281 del_gendisk(disk);
282 put_disk(disk);
283 blk_cleanup_queue(blk_dev->queue);
284 return rc;
285 }
286 }
287
288 set_capacity(disk, available_disk_size >> SECTOR_SHIFT);
289 revalidate_disk(disk);
290 return 0;
291 }
292
nd_blk_probe(struct device * dev)293 static int nd_blk_probe(struct device *dev)
294 {
295 struct nd_namespace_common *ndns;
296 struct nd_namespace_blk *nsblk;
297 struct nd_blk_device *blk_dev;
298 int rc;
299
300 ndns = nvdimm_namespace_common_probe(dev);
301 if (IS_ERR(ndns))
302 return PTR_ERR(ndns);
303
304 blk_dev = kzalloc(sizeof(*blk_dev), GFP_KERNEL);
305 if (!blk_dev)
306 return -ENOMEM;
307
308 nsblk = to_nd_namespace_blk(&ndns->dev);
309 blk_dev->disk_size = nvdimm_namespace_capacity(ndns);
310 blk_dev->ndbr = to_nd_blk_region(dev->parent);
311 blk_dev->nsblk = to_nd_namespace_blk(&ndns->dev);
312 blk_dev->internal_lbasize = roundup(nsblk->lbasize,
313 INT_LBASIZE_ALIGNMENT);
314 blk_dev->sector_size = ((nsblk->lbasize >= 4096) ? 4096 : 512);
315 dev_set_drvdata(dev, blk_dev);
316
317 ndns->rw_bytes = nd_blk_rw_bytes;
318 if (is_nd_btt(dev))
319 rc = nvdimm_namespace_attach_btt(ndns);
320 else if (nd_btt_probe(ndns, blk_dev) == 0) {
321 /* we'll come back as btt-blk */
322 rc = -ENXIO;
323 } else
324 rc = nd_blk_attach_disk(ndns, blk_dev);
325 if (rc)
326 kfree(blk_dev);
327 return rc;
328 }
329
nd_blk_detach_disk(struct nd_blk_device * blk_dev)330 static void nd_blk_detach_disk(struct nd_blk_device *blk_dev)
331 {
332 del_gendisk(blk_dev->disk);
333 put_disk(blk_dev->disk);
334 blk_cleanup_queue(blk_dev->queue);
335 }
336
nd_blk_remove(struct device * dev)337 static int nd_blk_remove(struct device *dev)
338 {
339 struct nd_blk_device *blk_dev = dev_get_drvdata(dev);
340
341 if (is_nd_btt(dev))
342 nvdimm_namespace_detach_btt(to_nd_btt(dev)->ndns);
343 else
344 nd_blk_detach_disk(blk_dev);
345 kfree(blk_dev);
346
347 return 0;
348 }
349
350 static struct nd_device_driver nd_blk_driver = {
351 .probe = nd_blk_probe,
352 .remove = nd_blk_remove,
353 .drv = {
354 .name = "nd_blk",
355 },
356 .type = ND_DRIVER_NAMESPACE_BLK,
357 };
358
nd_blk_init(void)359 static int __init nd_blk_init(void)
360 {
361 int rc;
362
363 rc = register_blkdev(0, "nd_blk");
364 if (rc < 0)
365 return rc;
366
367 nd_blk_major = rc;
368 rc = nd_driver_register(&nd_blk_driver);
369
370 if (rc < 0)
371 unregister_blkdev(nd_blk_major, "nd_blk");
372
373 return rc;
374 }
375
nd_blk_exit(void)376 static void __exit nd_blk_exit(void)
377 {
378 driver_unregister(&nd_blk_driver.drv);
379 unregister_blkdev(nd_blk_major, "nd_blk");
380 }
381
382 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
383 MODULE_LICENSE("GPL v2");
384 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_BLK);
385 module_init(nd_blk_init);
386 module_exit(nd_blk_exit);
387