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 
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 = lpp0b;
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 
51 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
52 {
53 	if (ns->bdev) {
54 		blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
55 		ns->bdev = NULL;
56 	}
57 }
58 
59 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns)
60 {
61 	struct blk_integrity *bi = bdev_get_integrity(ns->bdev);
62 
63 	if (bi) {
64 		ns->metadata_size = bi->tuple_size;
65 		if (bi->profile == &t10_pi_type1_crc)
66 			ns->pi_type = NVME_NS_DPS_PI_TYPE1;
67 		else if (bi->profile == &t10_pi_type3_crc)
68 			ns->pi_type = NVME_NS_DPS_PI_TYPE3;
69 		else
70 			/* Unsupported metadata type */
71 			ns->metadata_size = 0;
72 	}
73 }
74 
75 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
76 {
77 	int ret;
78 
79 	/*
80 	 * When buffered_io namespace attribute is enabled that means user want
81 	 * this block device to be used as a file, so block device can take
82 	 * an advantage of cache.
83 	 */
84 	if (ns->buffered_io)
85 		return -ENOTBLK;
86 
87 	ns->bdev = blkdev_get_by_path(ns->device_path,
88 			FMODE_READ | FMODE_WRITE, NULL);
89 	if (IS_ERR(ns->bdev)) {
90 		ret = PTR_ERR(ns->bdev);
91 		if (ret != -ENOTBLK) {
92 			pr_err("failed to open block device %s: (%ld)\n",
93 					ns->device_path, PTR_ERR(ns->bdev));
94 		}
95 		ns->bdev = NULL;
96 		return ret;
97 	}
98 	ns->size = bdev_nr_bytes(ns->bdev);
99 	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
100 
101 	ns->pi_type = 0;
102 	ns->metadata_size = 0;
103 	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY_T10))
104 		nvmet_bdev_ns_enable_integrity(ns);
105 
106 	if (bdev_is_zoned(ns->bdev)) {
107 		if (!nvmet_bdev_zns_enable(ns)) {
108 			nvmet_bdev_ns_disable(ns);
109 			return -EINVAL;
110 		}
111 		ns->csi = NVME_CSI_ZNS;
112 	}
113 
114 	return 0;
115 }
116 
117 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
118 {
119 	ns->size = bdev_nr_bytes(ns->bdev);
120 }
121 
122 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
123 {
124 	u16 status = NVME_SC_SUCCESS;
125 
126 	if (likely(blk_sts == BLK_STS_OK))
127 		return status;
128 	/*
129 	 * Right now there exists M : 1 mapping between block layer error
130 	 * to the NVMe status code (see nvme_error_status()). For consistency,
131 	 * when we reverse map we use most appropriate NVMe Status code from
132 	 * the group of the NVMe staus codes used in the nvme_error_status().
133 	 */
134 	switch (blk_sts) {
135 	case BLK_STS_NOSPC:
136 		status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
137 		req->error_loc = offsetof(struct nvme_rw_command, length);
138 		break;
139 	case BLK_STS_TARGET:
140 		status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
141 		req->error_loc = offsetof(struct nvme_rw_command, slba);
142 		break;
143 	case BLK_STS_NOTSUPP:
144 		req->error_loc = offsetof(struct nvme_common_command, opcode);
145 		switch (req->cmd->common.opcode) {
146 		case nvme_cmd_dsm:
147 		case nvme_cmd_write_zeroes:
148 			status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
149 			break;
150 		default:
151 			status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
152 		}
153 		break;
154 	case BLK_STS_MEDIUM:
155 		status = NVME_SC_ACCESS_DENIED;
156 		req->error_loc = offsetof(struct nvme_rw_command, nsid);
157 		break;
158 	case BLK_STS_IOERR:
159 	default:
160 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
161 		req->error_loc = offsetof(struct nvme_common_command, opcode);
162 	}
163 
164 	switch (req->cmd->common.opcode) {
165 	case nvme_cmd_read:
166 	case nvme_cmd_write:
167 		req->error_slba = le64_to_cpu(req->cmd->rw.slba);
168 		break;
169 	case nvme_cmd_write_zeroes:
170 		req->error_slba =
171 			le64_to_cpu(req->cmd->write_zeroes.slba);
172 		break;
173 	default:
174 		req->error_slba = 0;
175 	}
176 	return status;
177 }
178 
179 static void nvmet_bio_done(struct bio *bio)
180 {
181 	struct nvmet_req *req = bio->bi_private;
182 
183 	nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
184 	nvmet_req_bio_put(req, bio);
185 }
186 
187 #ifdef CONFIG_BLK_DEV_INTEGRITY
188 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
189 				struct sg_mapping_iter *miter)
190 {
191 	struct blk_integrity *bi;
192 	struct bio_integrity_payload *bip;
193 	int rc;
194 	size_t resid, len;
195 
196 	bi = bdev_get_integrity(req->ns->bdev);
197 	if (unlikely(!bi)) {
198 		pr_err("Unable to locate bio_integrity\n");
199 		return -ENODEV;
200 	}
201 
202 	bip = bio_integrity_alloc(bio, GFP_NOIO,
203 					bio_max_segs(req->metadata_sg_cnt));
204 	if (IS_ERR(bip)) {
205 		pr_err("Unable to allocate bio_integrity_payload\n");
206 		return PTR_ERR(bip);
207 	}
208 
209 	bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
210 	/* virtual start sector must be in integrity interval units */
211 	bip_set_seed(bip, bio->bi_iter.bi_sector >>
212 		     (bi->interval_exp - SECTOR_SHIFT));
213 
214 	resid = bip->bip_iter.bi_size;
215 	while (resid > 0 && sg_miter_next(miter)) {
216 		len = min_t(size_t, miter->length, resid);
217 		rc = bio_integrity_add_page(bio, miter->page, len,
218 					    offset_in_page(miter->addr));
219 		if (unlikely(rc != len)) {
220 			pr_err("bio_integrity_add_page() failed; %d\n", rc);
221 			sg_miter_stop(miter);
222 			return -ENOMEM;
223 		}
224 
225 		resid -= len;
226 		if (len < miter->length)
227 			miter->consumed -= miter->length - len;
228 	}
229 	sg_miter_stop(miter);
230 
231 	return 0;
232 }
233 #else
234 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
235 				struct sg_mapping_iter *miter)
236 {
237 	return -EINVAL;
238 }
239 #endif /* CONFIG_BLK_DEV_INTEGRITY */
240 
241 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
242 {
243 	unsigned int sg_cnt = req->sg_cnt;
244 	struct bio *bio;
245 	struct scatterlist *sg;
246 	struct blk_plug plug;
247 	sector_t sector;
248 	blk_opf_t opf;
249 	int i, rc;
250 	struct sg_mapping_iter prot_miter;
251 	unsigned int iter_flags;
252 	unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
253 
254 	if (!nvmet_check_transfer_len(req, total_len))
255 		return;
256 
257 	if (!req->sg_cnt) {
258 		nvmet_req_complete(req, 0);
259 		return;
260 	}
261 
262 	if (req->cmd->rw.opcode == nvme_cmd_write) {
263 		opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
264 		if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
265 			opf |= REQ_FUA;
266 		iter_flags = SG_MITER_TO_SG;
267 	} else {
268 		opf = REQ_OP_READ;
269 		iter_flags = SG_MITER_FROM_SG;
270 	}
271 
272 	if (is_pci_p2pdma_page(sg_page(req->sg)))
273 		opf |= REQ_NOMERGE;
274 
275 	sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
276 
277 	if (nvmet_use_inline_bvec(req)) {
278 		bio = &req->b.inline_bio;
279 		bio_init(bio, req->ns->bdev, req->inline_bvec,
280 			 ARRAY_SIZE(req->inline_bvec), opf);
281 	} else {
282 		bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf,
283 				GFP_KERNEL);
284 	}
285 	bio->bi_iter.bi_sector = sector;
286 	bio->bi_private = req;
287 	bio->bi_end_io = nvmet_bio_done;
288 
289 	blk_start_plug(&plug);
290 	if (req->metadata_len)
291 		sg_miter_start(&prot_miter, req->metadata_sg,
292 			       req->metadata_sg_cnt, iter_flags);
293 
294 	for_each_sg(req->sg, sg, req->sg_cnt, i) {
295 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
296 				!= sg->length) {
297 			struct bio *prev = bio;
298 
299 			if (req->metadata_len) {
300 				rc = nvmet_bdev_alloc_bip(req, bio,
301 							  &prot_miter);
302 				if (unlikely(rc)) {
303 					bio_io_error(bio);
304 					return;
305 				}
306 			}
307 
308 			bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt),
309 					opf, GFP_KERNEL);
310 			bio->bi_iter.bi_sector = sector;
311 
312 			bio_chain(bio, prev);
313 			submit_bio(prev);
314 		}
315 
316 		sector += sg->length >> 9;
317 		sg_cnt--;
318 	}
319 
320 	if (req->metadata_len) {
321 		rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
322 		if (unlikely(rc)) {
323 			bio_io_error(bio);
324 			return;
325 		}
326 	}
327 
328 	submit_bio(bio);
329 	blk_finish_plug(&plug);
330 }
331 
332 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
333 {
334 	struct bio *bio = &req->b.inline_bio;
335 
336 	if (!bdev_write_cache(req->ns->bdev)) {
337 		nvmet_req_complete(req, NVME_SC_SUCCESS);
338 		return;
339 	}
340 
341 	if (!nvmet_check_transfer_len(req, 0))
342 		return;
343 
344 	bio_init(bio, req->ns->bdev, req->inline_bvec,
345 		 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH);
346 	bio->bi_private = req;
347 	bio->bi_end_io = nvmet_bio_done;
348 
349 	submit_bio(bio);
350 }
351 
352 u16 nvmet_bdev_flush(struct nvmet_req *req)
353 {
354 	if (!bdev_write_cache(req->ns->bdev))
355 		return 0;
356 
357 	if (blkdev_issue_flush(req->ns->bdev))
358 		return NVME_SC_INTERNAL | NVME_SC_DNR;
359 	return 0;
360 }
361 
362 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
363 		struct nvme_dsm_range *range, struct bio **bio)
364 {
365 	struct nvmet_ns *ns = req->ns;
366 	int ret;
367 
368 	ret = __blkdev_issue_discard(ns->bdev,
369 			nvmet_lba_to_sect(ns, range->slba),
370 			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
371 			GFP_KERNEL, bio);
372 	if (ret && ret != -EOPNOTSUPP) {
373 		req->error_slba = le64_to_cpu(range->slba);
374 		return errno_to_nvme_status(req, ret);
375 	}
376 	return NVME_SC_SUCCESS;
377 }
378 
379 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
380 {
381 	struct nvme_dsm_range range;
382 	struct bio *bio = NULL;
383 	int i;
384 	u16 status;
385 
386 	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
387 		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
388 				sizeof(range));
389 		if (status)
390 			break;
391 
392 		status = nvmet_bdev_discard_range(req, &range, &bio);
393 		if (status)
394 			break;
395 	}
396 
397 	if (bio) {
398 		bio->bi_private = req;
399 		bio->bi_end_io = nvmet_bio_done;
400 		if (status)
401 			bio_io_error(bio);
402 		else
403 			submit_bio(bio);
404 	} else {
405 		nvmet_req_complete(req, status);
406 	}
407 }
408 
409 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
410 {
411 	if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
412 		return;
413 
414 	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
415 	case NVME_DSMGMT_AD:
416 		nvmet_bdev_execute_discard(req);
417 		return;
418 	case NVME_DSMGMT_IDR:
419 	case NVME_DSMGMT_IDW:
420 	default:
421 		/* Not supported yet */
422 		nvmet_req_complete(req, 0);
423 		return;
424 	}
425 }
426 
427 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
428 {
429 	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
430 	struct bio *bio = NULL;
431 	sector_t sector;
432 	sector_t nr_sector;
433 	int ret;
434 
435 	if (!nvmet_check_transfer_len(req, 0))
436 		return;
437 
438 	sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
439 	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
440 		(req->ns->blksize_shift - 9));
441 
442 	ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
443 			GFP_KERNEL, &bio, 0);
444 	if (bio) {
445 		bio->bi_private = req;
446 		bio->bi_end_io = nvmet_bio_done;
447 		submit_bio(bio);
448 	} else {
449 		nvmet_req_complete(req, errno_to_nvme_status(req, ret));
450 	}
451 }
452 
453 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
454 {
455 	switch (req->cmd->common.opcode) {
456 	case nvme_cmd_read:
457 	case nvme_cmd_write:
458 		req->execute = nvmet_bdev_execute_rw;
459 		if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
460 			req->metadata_len = nvmet_rw_metadata_len(req);
461 		return 0;
462 	case nvme_cmd_flush:
463 		req->execute = nvmet_bdev_execute_flush;
464 		return 0;
465 	case nvme_cmd_dsm:
466 		req->execute = nvmet_bdev_execute_dsm;
467 		return 0;
468 	case nvme_cmd_write_zeroes:
469 		req->execute = nvmet_bdev_execute_write_zeroes;
470 		return 0;
471 	default:
472 		return nvmet_report_invalid_opcode(req);
473 	}
474 }
475