1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * NVMe I/O command implementation.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5 */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/blkdev.h>
8 #include <linux/blk-integrity.h>
9 #include <linux/memremap.h>
10 #include <linux/module.h>
11 #include "nvmet.h"
12
nvmet_bdev_set_limits(struct block_device * bdev,struct nvme_id_ns * id)13 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)
14 {
15 /* Logical blocks per physical block, 0's based. */
16 const __le16 lpp0b = to0based(bdev_physical_block_size(bdev) /
17 bdev_logical_block_size(bdev));
18
19 /*
20 * For NVMe 1.2 and later, bit 1 indicates that the fields NAWUN,
21 * NAWUPF, and NACWU are defined for this namespace and should be
22 * used by the host for this namespace instead of the AWUN, AWUPF,
23 * and ACWU fields in the Identify Controller data structure. If
24 * any of these fields are zero that means that the corresponding
25 * field from the identify controller data structure should be used.
26 */
27 id->nsfeat |= 1 << 1;
28 id->nawun = lpp0b;
29 id->nawupf = lpp0b;
30 id->nacwu = lpp0b;
31
32 /*
33 * Bit 4 indicates that the fields NPWG, NPWA, NPDG, NPDA, and
34 * NOWS are defined for this namespace and should be used by
35 * the host for I/O optimization.
36 */
37 id->nsfeat |= 1 << 4;
38 /* NPWG = Namespace Preferred Write Granularity. 0's based */
39 id->npwg = to0based(bdev_io_min(bdev) / bdev_logical_block_size(bdev));
40 /* NPWA = Namespace Preferred Write Alignment. 0's based */
41 id->npwa = id->npwg;
42 /* NPDG = Namespace Preferred Deallocate Granularity. 0's based */
43 id->npdg = to0based(bdev_discard_granularity(bdev) /
44 bdev_logical_block_size(bdev));
45 /* NPDG = Namespace Preferred Deallocate Alignment */
46 id->npda = id->npdg;
47 /* NOWS = Namespace Optimal Write Size */
48 id->nows = to0based(bdev_io_opt(bdev) / bdev_logical_block_size(bdev));
49 }
50
nvmet_bdev_ns_disable(struct nvmet_ns * ns)51 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
52 {
53 if (ns->bdev_file) {
54 fput(ns->bdev_file);
55 ns->bdev = NULL;
56 ns->bdev_file = NULL;
57 }
58 }
59
nvmet_bdev_ns_enable_integrity(struct nvmet_ns * ns)60 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns)
61 {
62 struct blk_integrity *bi = bdev_get_integrity(ns->bdev);
63
64 if (!bi)
65 return;
66
67 if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC) {
68 ns->metadata_size = bi->tuple_size;
69 if (bi->flags & BLK_INTEGRITY_REF_TAG)
70 ns->pi_type = NVME_NS_DPS_PI_TYPE1;
71 else
72 ns->pi_type = NVME_NS_DPS_PI_TYPE3;
73 } else {
74 ns->metadata_size = 0;
75 }
76 }
77
nvmet_bdev_ns_enable(struct nvmet_ns * ns)78 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
79 {
80 int ret;
81
82 /*
83 * When buffered_io namespace attribute is enabled that means user want
84 * this block device to be used as a file, so block device can take
85 * an advantage of cache.
86 */
87 if (ns->buffered_io)
88 return -ENOTBLK;
89
90 ns->bdev_file = bdev_file_open_by_path(ns->device_path,
91 BLK_OPEN_READ | BLK_OPEN_WRITE, NULL, NULL);
92 if (IS_ERR(ns->bdev_file)) {
93 ret = PTR_ERR(ns->bdev_file);
94 if (ret != -ENOTBLK) {
95 pr_err("failed to open block device %s: (%d)\n",
96 ns->device_path, ret);
97 }
98 ns->bdev_file = NULL;
99 return ret;
100 }
101 ns->bdev = file_bdev(ns->bdev_file);
102 ns->size = bdev_nr_bytes(ns->bdev);
103 ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
104
105 ns->pi_type = 0;
106 ns->metadata_size = 0;
107 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
108 nvmet_bdev_ns_enable_integrity(ns);
109
110 if (bdev_is_zoned(ns->bdev)) {
111 if (!nvmet_bdev_zns_enable(ns)) {
112 nvmet_bdev_ns_disable(ns);
113 return -EINVAL;
114 }
115 ns->csi = NVME_CSI_ZNS;
116 }
117
118 return 0;
119 }
120
nvmet_bdev_ns_revalidate(struct nvmet_ns * ns)121 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
122 {
123 ns->size = bdev_nr_bytes(ns->bdev);
124 }
125
blk_to_nvme_status(struct nvmet_req * req,blk_status_t blk_sts)126 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
127 {
128 u16 status = NVME_SC_SUCCESS;
129
130 if (likely(blk_sts == BLK_STS_OK))
131 return status;
132 /*
133 * Right now there exists M : 1 mapping between block layer error
134 * to the NVMe status code (see nvme_error_status()). For consistency,
135 * when we reverse map we use most appropriate NVMe Status code from
136 * the group of the NVMe status codes used in the nvme_error_status().
137 */
138 switch (blk_sts) {
139 case BLK_STS_NOSPC:
140 status = NVME_SC_CAP_EXCEEDED | NVME_STATUS_DNR;
141 req->error_loc = offsetof(struct nvme_rw_command, length);
142 break;
143 case BLK_STS_TARGET:
144 status = NVME_SC_LBA_RANGE | NVME_STATUS_DNR;
145 req->error_loc = offsetof(struct nvme_rw_command, slba);
146 break;
147 case BLK_STS_NOTSUPP:
148 status = NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR;
149 req->error_loc = offsetof(struct nvme_common_command, opcode);
150 break;
151 case BLK_STS_MEDIUM:
152 status = NVME_SC_ACCESS_DENIED;
153 req->error_loc = offsetof(struct nvme_rw_command, nsid);
154 break;
155 case BLK_STS_IOERR:
156 default:
157 status = NVME_SC_INTERNAL | NVME_STATUS_DNR;
158 req->error_loc = offsetof(struct nvme_common_command, opcode);
159 }
160
161 switch (req->cmd->common.opcode) {
162 case nvme_cmd_read:
163 case nvme_cmd_write:
164 req->error_slba = le64_to_cpu(req->cmd->rw.slba);
165 break;
166 case nvme_cmd_write_zeroes:
167 req->error_slba =
168 le64_to_cpu(req->cmd->write_zeroes.slba);
169 break;
170 default:
171 req->error_slba = 0;
172 }
173 return status;
174 }
175
nvmet_bio_done(struct bio * bio)176 static void nvmet_bio_done(struct bio *bio)
177 {
178 struct nvmet_req *req = bio->bi_private;
179
180 nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
181 nvmet_req_bio_put(req, bio);
182 }
183
184 #ifdef CONFIG_BLK_DEV_INTEGRITY
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)185 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
186 struct sg_mapping_iter *miter)
187 {
188 struct blk_integrity *bi;
189 struct bio_integrity_payload *bip;
190 int rc;
191 size_t resid, len;
192
193 bi = bdev_get_integrity(req->ns->bdev);
194 if (unlikely(!bi)) {
195 pr_err("Unable to locate bio_integrity\n");
196 return -ENODEV;
197 }
198
199 bip = bio_integrity_alloc(bio, GFP_NOIO,
200 bio_max_segs(req->metadata_sg_cnt));
201 if (IS_ERR(bip)) {
202 pr_err("Unable to allocate bio_integrity_payload\n");
203 return PTR_ERR(bip);
204 }
205
206 /* virtual start sector must be in integrity interval units */
207 bip_set_seed(bip, bio->bi_iter.bi_sector >>
208 (bi->interval_exp - SECTOR_SHIFT));
209
210 resid = bio_integrity_bytes(bi, bio_sectors(bio));
211 while (resid > 0 && sg_miter_next(miter)) {
212 len = min_t(size_t, miter->length, resid);
213 rc = bio_integrity_add_page(bio, miter->page, len,
214 offset_in_page(miter->addr));
215 if (unlikely(rc != len)) {
216 pr_err("bio_integrity_add_page() failed; %d\n", rc);
217 sg_miter_stop(miter);
218 return -ENOMEM;
219 }
220
221 resid -= len;
222 if (len < miter->length)
223 miter->consumed -= miter->length - len;
224 }
225 sg_miter_stop(miter);
226
227 return 0;
228 }
229 #else
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)230 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
231 struct sg_mapping_iter *miter)
232 {
233 return -EINVAL;
234 }
235 #endif /* CONFIG_BLK_DEV_INTEGRITY */
236
nvmet_bdev_execute_rw(struct nvmet_req * req)237 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
238 {
239 unsigned int sg_cnt = req->sg_cnt;
240 struct bio *bio;
241 struct scatterlist *sg;
242 struct blk_plug plug;
243 sector_t sector;
244 blk_opf_t opf;
245 int i, rc;
246 struct sg_mapping_iter prot_miter;
247 unsigned int iter_flags;
248 unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
249
250 if (!nvmet_check_transfer_len(req, total_len))
251 return;
252
253 if (!req->sg_cnt) {
254 nvmet_req_complete(req, 0);
255 return;
256 }
257
258 if (req->cmd->rw.opcode == nvme_cmd_write) {
259 opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
260 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
261 opf |= REQ_FUA;
262 iter_flags = SG_MITER_TO_SG;
263 } else {
264 opf = REQ_OP_READ;
265 iter_flags = SG_MITER_FROM_SG;
266 }
267
268 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_LR))
269 opf |= REQ_FAILFAST_DEV;
270
271 if (is_pci_p2pdma_page(sg_page(req->sg)))
272 opf |= REQ_NOMERGE;
273
274 sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
275
276 if (nvmet_use_inline_bvec(req)) {
277 bio = &req->b.inline_bio;
278 bio_init(bio, req->ns->bdev, req->inline_bvec,
279 ARRAY_SIZE(req->inline_bvec), opf);
280 } else {
281 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf,
282 GFP_KERNEL);
283 }
284 bio->bi_iter.bi_sector = sector;
285 bio->bi_private = req;
286 bio->bi_end_io = nvmet_bio_done;
287
288 blk_start_plug(&plug);
289 if (req->metadata_len)
290 sg_miter_start(&prot_miter, req->metadata_sg,
291 req->metadata_sg_cnt, iter_flags);
292
293 for_each_sg(req->sg, sg, req->sg_cnt, i) {
294 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
295 != sg->length) {
296 struct bio *prev = bio;
297
298 if (req->metadata_len) {
299 rc = nvmet_bdev_alloc_bip(req, bio,
300 &prot_miter);
301 if (unlikely(rc)) {
302 bio_io_error(bio);
303 return;
304 }
305 }
306
307 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt),
308 opf, GFP_KERNEL);
309 bio->bi_iter.bi_sector = sector;
310
311 bio_chain(bio, prev);
312 submit_bio(prev);
313 }
314
315 sector += sg->length >> 9;
316 sg_cnt--;
317 }
318
319 if (req->metadata_len) {
320 rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
321 if (unlikely(rc)) {
322 bio_io_error(bio);
323 return;
324 }
325 }
326
327 submit_bio(bio);
328 blk_finish_plug(&plug);
329 }
330
nvmet_bdev_execute_flush(struct nvmet_req * req)331 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
332 {
333 struct bio *bio = &req->b.inline_bio;
334
335 if (!bdev_write_cache(req->ns->bdev)) {
336 nvmet_req_complete(req, NVME_SC_SUCCESS);
337 return;
338 }
339
340 if (!nvmet_check_transfer_len(req, 0))
341 return;
342
343 bio_init(bio, req->ns->bdev, req->inline_bvec,
344 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH);
345 bio->bi_private = req;
346 bio->bi_end_io = nvmet_bio_done;
347
348 submit_bio(bio);
349 }
350
nvmet_bdev_flush(struct nvmet_req * req)351 u16 nvmet_bdev_flush(struct nvmet_req *req)
352 {
353 if (!bdev_write_cache(req->ns->bdev))
354 return 0;
355
356 if (blkdev_issue_flush(req->ns->bdev))
357 return NVME_SC_INTERNAL | NVME_STATUS_DNR;
358 return 0;
359 }
360
nvmet_bdev_discard_range(struct nvmet_req * req,struct nvme_dsm_range * range,struct bio ** bio)361 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
362 struct nvme_dsm_range *range, struct bio **bio)
363 {
364 struct nvmet_ns *ns = req->ns;
365 int ret;
366
367 ret = __blkdev_issue_discard(ns->bdev,
368 nvmet_lba_to_sect(ns, range->slba),
369 le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
370 GFP_KERNEL, bio);
371 if (ret && ret != -EOPNOTSUPP) {
372 req->error_slba = le64_to_cpu(range->slba);
373 return errno_to_nvme_status(req, ret);
374 }
375 return NVME_SC_SUCCESS;
376 }
377
nvmet_bdev_execute_discard(struct nvmet_req * req)378 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
379 {
380 struct nvme_dsm_range range;
381 struct bio *bio = NULL;
382 int i;
383 u16 status;
384
385 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
386 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
387 sizeof(range));
388 if (status)
389 break;
390
391 status = nvmet_bdev_discard_range(req, &range, &bio);
392 if (status)
393 break;
394 }
395
396 if (bio) {
397 bio->bi_private = req;
398 bio->bi_end_io = nvmet_bio_done;
399 if (status)
400 bio_io_error(bio);
401 else
402 submit_bio(bio);
403 } else {
404 nvmet_req_complete(req, status);
405 }
406 }
407
nvmet_bdev_execute_dsm(struct nvmet_req * req)408 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
409 {
410 if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
411 return;
412
413 switch (le32_to_cpu(req->cmd->dsm.attributes)) {
414 case NVME_DSMGMT_AD:
415 nvmet_bdev_execute_discard(req);
416 return;
417 case NVME_DSMGMT_IDR:
418 case NVME_DSMGMT_IDW:
419 default:
420 /* Not supported yet */
421 nvmet_req_complete(req, 0);
422 return;
423 }
424 }
425
nvmet_bdev_execute_write_zeroes(struct nvmet_req * req)426 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
427 {
428 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
429 struct bio *bio = NULL;
430 sector_t sector;
431 sector_t nr_sector;
432 int ret;
433
434 if (!nvmet_check_transfer_len(req, 0))
435 return;
436
437 sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
438 nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
439 (req->ns->blksize_shift - 9));
440
441 ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
442 GFP_KERNEL, &bio, 0);
443 if (bio) {
444 bio->bi_private = req;
445 bio->bi_end_io = nvmet_bio_done;
446 submit_bio(bio);
447 } else {
448 nvmet_req_complete(req, errno_to_nvme_status(req, ret));
449 }
450 }
451
nvmet_bdev_parse_io_cmd(struct nvmet_req * req)452 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
453 {
454 switch (req->cmd->common.opcode) {
455 case nvme_cmd_read:
456 case nvme_cmd_write:
457 req->execute = nvmet_bdev_execute_rw;
458 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
459 req->metadata_len = nvmet_rw_metadata_len(req);
460 return 0;
461 case nvme_cmd_flush:
462 req->execute = nvmet_bdev_execute_flush;
463 return 0;
464 case nvme_cmd_dsm:
465 req->execute = nvmet_bdev_execute_dsm;
466 return 0;
467 case nvme_cmd_write_zeroes:
468 req->execute = nvmet_bdev_execute_write_zeroes;
469 return 0;
470 default:
471 return nvmet_report_invalid_opcode(req);
472 }
473 }
474