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 /* Set WZDS and DRB if device supports unmapped write zeroes */
51 if (bdev_write_zeroes_unmap_sectors(bdev))
52 id->dlfeat = (1 << 3) | 0x1;
53 }
54
nvmet_bdev_ns_disable(struct nvmet_ns * ns)55 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
56 {
57 if (ns->bdev_file) {
58 fput(ns->bdev_file);
59 ns->bdev = NULL;
60 ns->bdev_file = NULL;
61 }
62 }
63
nvmet_bdev_ns_enable_integrity(struct nvmet_ns * ns)64 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns)
65 {
66 struct blk_integrity *bi = bdev_get_integrity(ns->bdev);
67
68 if (!bi)
69 return;
70
71 if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC) {
72 ns->metadata_size = bi->metadata_size;
73 if (bi->flags & BLK_INTEGRITY_REF_TAG)
74 ns->pi_type = NVME_NS_DPS_PI_TYPE1;
75 else
76 ns->pi_type = NVME_NS_DPS_PI_TYPE3;
77 } else {
78 ns->metadata_size = 0;
79 }
80 }
81
nvmet_bdev_ns_enable(struct nvmet_ns * ns)82 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
83 {
84 int ret;
85
86 /*
87 * When buffered_io namespace attribute is enabled that means user want
88 * this block device to be used as a file, so block device can take
89 * an advantage of cache.
90 */
91 if (ns->buffered_io)
92 return -ENOTBLK;
93
94 ns->bdev_file = bdev_file_open_by_path(ns->device_path,
95 BLK_OPEN_READ | BLK_OPEN_WRITE, NULL, NULL);
96 if (IS_ERR(ns->bdev_file)) {
97 ret = PTR_ERR(ns->bdev_file);
98 if (ret != -ENOTBLK) {
99 pr_err("failed to open block device %s: (%d)\n",
100 ns->device_path, ret);
101 }
102 ns->bdev_file = NULL;
103 return ret;
104 }
105 ns->bdev = file_bdev(ns->bdev_file);
106 ns->size = bdev_nr_bytes(ns->bdev);
107 ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
108
109 ns->pi_type = 0;
110 ns->metadata_size = 0;
111 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
112 nvmet_bdev_ns_enable_integrity(ns);
113
114 if (bdev_is_zoned(ns->bdev)) {
115 if (!nvmet_bdev_zns_enable(ns)) {
116 nvmet_bdev_ns_disable(ns);
117 return -EINVAL;
118 }
119 ns->csi = NVME_CSI_ZNS;
120 }
121
122 return 0;
123 }
124
nvmet_bdev_ns_revalidate(struct nvmet_ns * ns)125 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
126 {
127 ns->size = bdev_nr_bytes(ns->bdev);
128 }
129
blk_to_nvme_status(struct nvmet_req * req,blk_status_t blk_sts)130 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
131 {
132 u16 status = NVME_SC_SUCCESS;
133
134 if (likely(blk_sts == BLK_STS_OK))
135 return status;
136 /*
137 * Right now there exists M : 1 mapping between block layer error
138 * to the NVMe status code (see nvme_error_status()). For consistency,
139 * when we reverse map we use most appropriate NVMe Status code from
140 * the group of the NVMe status codes used in the nvme_error_status().
141 */
142 switch (blk_sts) {
143 case BLK_STS_NOSPC:
144 status = NVME_SC_CAP_EXCEEDED | NVME_STATUS_DNR;
145 req->error_loc = offsetof(struct nvme_rw_command, length);
146 break;
147 case BLK_STS_TARGET:
148 status = NVME_SC_LBA_RANGE | NVME_STATUS_DNR;
149 req->error_loc = offsetof(struct nvme_rw_command, slba);
150 break;
151 case BLK_STS_NOTSUPP:
152 status = NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR;
153 req->error_loc = offsetof(struct nvme_common_command, opcode);
154 break;
155 case BLK_STS_MEDIUM:
156 status = NVME_SC_ACCESS_DENIED;
157 req->error_loc = offsetof(struct nvme_rw_command, nsid);
158 break;
159 case BLK_STS_IOERR:
160 default:
161 status = NVME_SC_INTERNAL | NVME_STATUS_DNR;
162 req->error_loc = offsetof(struct nvme_common_command, opcode);
163 }
164
165 switch (req->cmd->common.opcode) {
166 case nvme_cmd_read:
167 case nvme_cmd_write:
168 req->error_slba = le64_to_cpu(req->cmd->rw.slba);
169 break;
170 case nvme_cmd_write_zeroes:
171 req->error_slba =
172 le64_to_cpu(req->cmd->write_zeroes.slba);
173 break;
174 default:
175 req->error_slba = 0;
176 }
177 return status;
178 }
179
nvmet_bio_done(struct bio * bio)180 static void nvmet_bio_done(struct bio *bio)
181 {
182 struct nvmet_req *req = bio->bi_private;
183 blk_status_t blk_status = bio->bi_status;
184
185 nvmet_req_bio_put(req, bio);
186 nvmet_req_complete(req, blk_to_nvme_status(req, blk_status));
187 }
188
189 #ifdef CONFIG_BLK_DEV_INTEGRITY
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)190 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
191 struct sg_mapping_iter *miter)
192 {
193 struct blk_integrity *bi;
194 struct bio_integrity_payload *bip;
195 int rc;
196 size_t resid, len;
197
198 bi = bdev_get_integrity(req->ns->bdev);
199 if (unlikely(!bi)) {
200 pr_err("Unable to locate bio_integrity\n");
201 return -ENODEV;
202 }
203
204 bip = bio_integrity_alloc(bio, GFP_NOIO,
205 bio_max_segs(req->metadata_sg_cnt));
206 if (IS_ERR(bip)) {
207 pr_err("Unable to allocate bio_integrity_payload\n");
208 return PTR_ERR(bip);
209 }
210
211 /* virtual start sector must be in integrity interval units */
212 bip_set_seed(bip, bio->bi_iter.bi_sector >>
213 (bi->interval_exp - SECTOR_SHIFT));
214
215 resid = bio_integrity_bytes(bi, bio_sectors(bio));
216 while (resid > 0 && sg_miter_next(miter)) {
217 len = min_t(size_t, miter->length, resid);
218 rc = bio_integrity_add_page(bio, miter->page, len,
219 offset_in_page(miter->addr));
220 if (unlikely(rc != len)) {
221 pr_err("bio_integrity_add_page() failed; %d\n", rc);
222 sg_miter_stop(miter);
223 return -ENOMEM;
224 }
225
226 resid -= len;
227 if (len < miter->length)
228 miter->consumed -= miter->length - len;
229 }
230 sg_miter_stop(miter);
231
232 return 0;
233 }
234 #else
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)235 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
236 struct sg_mapping_iter *miter)
237 {
238 return -EINVAL;
239 }
240 #endif /* CONFIG_BLK_DEV_INTEGRITY */
241
nvmet_bdev_execute_rw(struct nvmet_req * req)242 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
243 {
244 unsigned int sg_cnt = req->sg_cnt;
245 struct bio *bio;
246 struct scatterlist *sg;
247 struct blk_plug plug;
248 sector_t sector;
249 blk_opf_t opf;
250 int i, rc;
251 struct sg_mapping_iter prot_miter;
252 unsigned int iter_flags;
253 unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
254
255 if (!nvmet_check_transfer_len(req, total_len))
256 return;
257
258 if (!req->sg_cnt) {
259 nvmet_req_complete(req, 0);
260 return;
261 }
262
263 if (req->cmd->rw.opcode == nvme_cmd_write) {
264 opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
265 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
266 opf |= REQ_FUA;
267 iter_flags = SG_MITER_TO_SG;
268 } else {
269 opf = REQ_OP_READ;
270 iter_flags = SG_MITER_FROM_SG;
271 }
272
273 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_LR))
274 opf |= REQ_FAILFAST_DEV;
275
276 if (is_pci_p2pdma_page(sg_page(req->sg)))
277 opf |= REQ_NOMERGE;
278
279 sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
280
281 if (nvmet_use_inline_bvec(req)) {
282 bio = &req->b.inline_bio;
283 bio_init(bio, req->ns->bdev, req->inline_bvec,
284 ARRAY_SIZE(req->inline_bvec), opf);
285 } else {
286 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf,
287 GFP_KERNEL);
288 }
289 bio->bi_iter.bi_sector = sector;
290 bio->bi_private = req;
291 bio->bi_end_io = nvmet_bio_done;
292
293 blk_start_plug(&plug);
294 if (req->metadata_len)
295 sg_miter_start(&prot_miter, req->metadata_sg,
296 req->metadata_sg_cnt, iter_flags);
297
298 for_each_sg(req->sg, sg, req->sg_cnt, i) {
299 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
300 != sg->length) {
301 struct bio *prev = bio;
302
303 if (req->metadata_len) {
304 rc = nvmet_bdev_alloc_bip(req, bio,
305 &prot_miter);
306 if (unlikely(rc)) {
307 bio_io_error(bio);
308 return;
309 }
310 }
311
312 bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt),
313 opf, GFP_KERNEL);
314 bio->bi_iter.bi_sector = sector;
315
316 bio_chain(bio, prev);
317 submit_bio(prev);
318 }
319
320 sector += sg->length >> 9;
321 sg_cnt--;
322 }
323
324 if (req->metadata_len) {
325 rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
326 if (unlikely(rc)) {
327 bio_io_error(bio);
328 return;
329 }
330 }
331
332 submit_bio(bio);
333 blk_finish_plug(&plug);
334 }
335
nvmet_bdev_execute_flush(struct nvmet_req * req)336 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
337 {
338 struct bio *bio = &req->b.inline_bio;
339
340 if (!bdev_write_cache(req->ns->bdev)) {
341 nvmet_req_complete(req, NVME_SC_SUCCESS);
342 return;
343 }
344
345 if (!nvmet_check_transfer_len(req, 0))
346 return;
347
348 bio_init(bio, req->ns->bdev, req->inline_bvec,
349 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH);
350 bio->bi_private = req;
351 bio->bi_end_io = nvmet_bio_done;
352
353 submit_bio(bio);
354 }
355
nvmet_bdev_flush(struct nvmet_req * req)356 u16 nvmet_bdev_flush(struct nvmet_req *req)
357 {
358 if (!bdev_write_cache(req->ns->bdev))
359 return 0;
360
361 if (blkdev_issue_flush(req->ns->bdev))
362 return NVME_SC_INTERNAL | NVME_STATUS_DNR;
363 return 0;
364 }
365
nvmet_bdev_discard_range(struct nvmet_req * req,struct nvme_dsm_range * range,struct bio ** bio)366 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
367 struct nvme_dsm_range *range, struct bio **bio)
368 {
369 struct nvmet_ns *ns = req->ns;
370 int ret;
371
372 ret = __blkdev_issue_discard(ns->bdev,
373 nvmet_lba_to_sect(ns, range->slba),
374 le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
375 GFP_KERNEL, bio);
376 if (ret && ret != -EOPNOTSUPP) {
377 req->error_slba = le64_to_cpu(range->slba);
378 return errno_to_nvme_status(req, ret);
379 }
380 return NVME_SC_SUCCESS;
381 }
382
nvmet_bdev_execute_discard(struct nvmet_req * req)383 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
384 {
385 struct nvme_dsm_range range;
386 struct bio *bio = NULL;
387 int i;
388 u16 status;
389
390 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
391 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
392 sizeof(range));
393 if (status)
394 break;
395
396 status = nvmet_bdev_discard_range(req, &range, &bio);
397 if (status)
398 break;
399 }
400
401 if (bio) {
402 bio->bi_private = req;
403 bio->bi_end_io = nvmet_bio_done;
404 if (status)
405 bio_io_error(bio);
406 else
407 submit_bio(bio);
408 } else {
409 nvmet_req_complete(req, status);
410 }
411 }
412
nvmet_bdev_execute_dsm(struct nvmet_req * req)413 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
414 {
415 if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
416 return;
417
418 switch (le32_to_cpu(req->cmd->dsm.attributes)) {
419 case NVME_DSMGMT_AD:
420 nvmet_bdev_execute_discard(req);
421 return;
422 case NVME_DSMGMT_IDR:
423 case NVME_DSMGMT_IDW:
424 default:
425 /* Not supported yet */
426 nvmet_req_complete(req, 0);
427 return;
428 }
429 }
430
nvmet_bdev_execute_write_zeroes(struct nvmet_req * req)431 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
432 {
433 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
434 struct bio *bio = NULL;
435 sector_t sector;
436 sector_t nr_sector;
437 int ret;
438
439 if (!nvmet_check_transfer_len(req, 0))
440 return;
441
442 sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
443 nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
444 (req->ns->blksize_shift - 9));
445
446 ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
447 GFP_KERNEL, &bio, 0);
448 if (bio) {
449 bio->bi_private = req;
450 bio->bi_end_io = nvmet_bio_done;
451 submit_bio(bio);
452 } else {
453 nvmet_req_complete(req, errno_to_nvme_status(req, ret));
454 }
455 }
456
nvmet_bdev_parse_io_cmd(struct nvmet_req * req)457 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
458 {
459 switch (req->cmd->common.opcode) {
460 case nvme_cmd_read:
461 case nvme_cmd_write:
462 req->execute = nvmet_bdev_execute_rw;
463 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
464 req->metadata_len = nvmet_rw_metadata_len(req);
465 return 0;
466 case nvme_cmd_flush:
467 req->execute = nvmet_bdev_execute_flush;
468 return 0;
469 case nvme_cmd_dsm:
470 req->execute = nvmet_bdev_execute_dsm;
471 return 0;
472 case nvme_cmd_write_zeroes:
473 req->execute = nvmet_bdev_execute_write_zeroes;
474 return 0;
475 default:
476 return nvmet_report_invalid_opcode(req);
477 }
478 }
479