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 	const struct queue_limits *ql = &bdev_get_queue(bdev)->limits;
16 	/* Number of logical blocks per physical block. */
17 	const u32 lpp = ql->physical_block_size / ql->logical_block_size;
18 	/* Logical blocks per physical block, 0's based. */
19 	const __le16 lpp0b = to0based(lpp);
20 
21 	/*
22 	 * For NVMe 1.2 and later, bit 1 indicates that the fields NAWUN,
23 	 * NAWUPF, and NACWU are defined for this namespace and should be
24 	 * used by the host for this namespace instead of the AWUN, AWUPF,
25 	 * and ACWU fields in the Identify Controller data structure. If
26 	 * any of these fields are zero that means that the corresponding
27 	 * field from the identify controller data structure should be used.
28 	 */
29 	id->nsfeat |= 1 << 1;
30 	id->nawun = lpp0b;
31 	id->nawupf = lpp0b;
32 	id->nacwu = lpp0b;
33 
34 	/*
35 	 * Bit 4 indicates that the fields NPWG, NPWA, NPDG, NPDA, and
36 	 * NOWS are defined for this namespace and should be used by
37 	 * the host for I/O optimization.
38 	 */
39 	id->nsfeat |= 1 << 4;
40 	/* NPWG = Namespace Preferred Write Granularity. 0's based */
41 	id->npwg = lpp0b;
42 	/* NPWA = Namespace Preferred Write Alignment. 0's based */
43 	id->npwa = id->npwg;
44 	/* NPDG = Namespace Preferred Deallocate Granularity. 0's based */
45 	id->npdg = to0based(ql->discard_granularity / ql->logical_block_size);
46 	/* NPDG = Namespace Preferred Deallocate Alignment */
47 	id->npda = id->npdg;
48 	/* NOWS = Namespace Optimal Write Size */
49 	id->nows = to0based(ql->io_opt / ql->logical_block_size);
50 }
51 
52 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
53 {
54 	if (ns->bdev) {
55 		blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
56 		ns->bdev = NULL;
57 	}
58 }
59 
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 		ns->metadata_size = bi->tuple_size;
66 		if (bi->profile == &t10_pi_type1_crc)
67 			ns->pi_type = NVME_NS_DPS_PI_TYPE1;
68 		else if (bi->profile == &t10_pi_type3_crc)
69 			ns->pi_type = NVME_NS_DPS_PI_TYPE3;
70 		else
71 			/* Unsupported metadata type */
72 			ns->metadata_size = 0;
73 	}
74 }
75 
76 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
77 {
78 	int ret;
79 
80 	/*
81 	 * When buffered_io namespace attribute is enabled that means user want
82 	 * this block device to be used as a file, so block device can take
83 	 * an advantage of cache.
84 	 */
85 	if (ns->buffered_io)
86 		return -ENOTBLK;
87 
88 	ns->bdev = blkdev_get_by_path(ns->device_path,
89 			FMODE_READ | FMODE_WRITE, NULL);
90 	if (IS_ERR(ns->bdev)) {
91 		ret = PTR_ERR(ns->bdev);
92 		if (ret != -ENOTBLK) {
93 			pr_err("failed to open block device %s: (%ld)\n",
94 					ns->device_path, PTR_ERR(ns->bdev));
95 		}
96 		ns->bdev = NULL;
97 		return ret;
98 	}
99 	ns->size = bdev_nr_bytes(ns->bdev);
100 	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
101 
102 	ns->pi_type = 0;
103 	ns->metadata_size = 0;
104 	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY_T10))
105 		nvmet_bdev_ns_enable_integrity(ns);
106 
107 	if (bdev_is_zoned(ns->bdev)) {
108 		if (!nvmet_bdev_zns_enable(ns)) {
109 			nvmet_bdev_ns_disable(ns);
110 			return -EINVAL;
111 		}
112 		ns->csi = NVME_CSI_ZNS;
113 	}
114 
115 	return 0;
116 }
117 
118 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
119 {
120 	ns->size = bdev_nr_bytes(ns->bdev);
121 }
122 
123 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
124 {
125 	u16 status = NVME_SC_SUCCESS;
126 
127 	if (likely(blk_sts == BLK_STS_OK))
128 		return status;
129 	/*
130 	 * Right now there exists M : 1 mapping between block layer error
131 	 * to the NVMe status code (see nvme_error_status()). For consistency,
132 	 * when we reverse map we use most appropriate NVMe Status code from
133 	 * the group of the NVMe staus codes used in the nvme_error_status().
134 	 */
135 	switch (blk_sts) {
136 	case BLK_STS_NOSPC:
137 		status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
138 		req->error_loc = offsetof(struct nvme_rw_command, length);
139 		break;
140 	case BLK_STS_TARGET:
141 		status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
142 		req->error_loc = offsetof(struct nvme_rw_command, slba);
143 		break;
144 	case BLK_STS_NOTSUPP:
145 		req->error_loc = offsetof(struct nvme_common_command, opcode);
146 		switch (req->cmd->common.opcode) {
147 		case nvme_cmd_dsm:
148 		case nvme_cmd_write_zeroes:
149 			status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
150 			break;
151 		default:
152 			status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
153 		}
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_SC_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 
180 static void nvmet_bio_done(struct bio *bio)
181 {
182 	struct nvmet_req *req = bio->bi_private;
183 
184 	nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
185 	nvmet_req_bio_put(req, bio);
186 }
187 
188 #ifdef CONFIG_BLK_DEV_INTEGRITY
189 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
190 				struct sg_mapping_iter *miter)
191 {
192 	struct blk_integrity *bi;
193 	struct bio_integrity_payload *bip;
194 	int rc;
195 	size_t resid, len;
196 
197 	bi = bdev_get_integrity(req->ns->bdev);
198 	if (unlikely(!bi)) {
199 		pr_err("Unable to locate bio_integrity\n");
200 		return -ENODEV;
201 	}
202 
203 	bip = bio_integrity_alloc(bio, GFP_NOIO,
204 					bio_max_segs(req->metadata_sg_cnt));
205 	if (IS_ERR(bip)) {
206 		pr_err("Unable to allocate bio_integrity_payload\n");
207 		return PTR_ERR(bip);
208 	}
209 
210 	bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
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 = bip->bip_iter.bi_size;
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
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 
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 (is_pci_p2pdma_page(sg_page(req->sg)))
274 		opf |= REQ_NOMERGE;
275 
276 	sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
277 
278 	if (nvmet_use_inline_bvec(req)) {
279 		bio = &req->b.inline_bio;
280 		bio_init(bio, req->ns->bdev, req->inline_bvec,
281 			 ARRAY_SIZE(req->inline_bvec), opf);
282 	} else {
283 		bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt), opf,
284 				GFP_KERNEL);
285 	}
286 	bio->bi_iter.bi_sector = sector;
287 	bio->bi_private = req;
288 	bio->bi_end_io = nvmet_bio_done;
289 
290 	blk_start_plug(&plug);
291 	if (req->metadata_len)
292 		sg_miter_start(&prot_miter, req->metadata_sg,
293 			       req->metadata_sg_cnt, iter_flags);
294 
295 	for_each_sg(req->sg, sg, req->sg_cnt, i) {
296 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
297 				!= sg->length) {
298 			struct bio *prev = bio;
299 
300 			if (req->metadata_len) {
301 				rc = nvmet_bdev_alloc_bip(req, bio,
302 							  &prot_miter);
303 				if (unlikely(rc)) {
304 					bio_io_error(bio);
305 					return;
306 				}
307 			}
308 
309 			bio = bio_alloc(req->ns->bdev, bio_max_segs(sg_cnt),
310 					opf, GFP_KERNEL);
311 			bio->bi_iter.bi_sector = sector;
312 
313 			bio_chain(bio, prev);
314 			submit_bio(prev);
315 		}
316 
317 		sector += sg->length >> 9;
318 		sg_cnt--;
319 	}
320 
321 	if (req->metadata_len) {
322 		rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
323 		if (unlikely(rc)) {
324 			bio_io_error(bio);
325 			return;
326 		}
327 	}
328 
329 	submit_bio(bio);
330 	blk_finish_plug(&plug);
331 }
332 
333 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
334 {
335 	struct bio *bio = &req->b.inline_bio;
336 
337 	if (!nvmet_check_transfer_len(req, 0))
338 		return;
339 
340 	bio_init(bio, req->ns->bdev, req->inline_bvec,
341 		 ARRAY_SIZE(req->inline_bvec), REQ_OP_WRITE | REQ_PREFLUSH);
342 	bio->bi_private = req;
343 	bio->bi_end_io = nvmet_bio_done;
344 
345 	submit_bio(bio);
346 }
347 
348 u16 nvmet_bdev_flush(struct nvmet_req *req)
349 {
350 	if (blkdev_issue_flush(req->ns->bdev))
351 		return NVME_SC_INTERNAL | NVME_SC_DNR;
352 	return 0;
353 }
354 
355 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
356 		struct nvme_dsm_range *range, struct bio **bio)
357 {
358 	struct nvmet_ns *ns = req->ns;
359 	int ret;
360 
361 	ret = __blkdev_issue_discard(ns->bdev,
362 			nvmet_lba_to_sect(ns, range->slba),
363 			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
364 			GFP_KERNEL, bio);
365 	if (ret && ret != -EOPNOTSUPP) {
366 		req->error_slba = le64_to_cpu(range->slba);
367 		return errno_to_nvme_status(req, ret);
368 	}
369 	return NVME_SC_SUCCESS;
370 }
371 
372 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
373 {
374 	struct nvme_dsm_range range;
375 	struct bio *bio = NULL;
376 	int i;
377 	u16 status;
378 
379 	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
380 		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
381 				sizeof(range));
382 		if (status)
383 			break;
384 
385 		status = nvmet_bdev_discard_range(req, &range, &bio);
386 		if (status)
387 			break;
388 	}
389 
390 	if (bio) {
391 		bio->bi_private = req;
392 		bio->bi_end_io = nvmet_bio_done;
393 		if (status)
394 			bio_io_error(bio);
395 		else
396 			submit_bio(bio);
397 	} else {
398 		nvmet_req_complete(req, status);
399 	}
400 }
401 
402 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
403 {
404 	if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
405 		return;
406 
407 	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
408 	case NVME_DSMGMT_AD:
409 		nvmet_bdev_execute_discard(req);
410 		return;
411 	case NVME_DSMGMT_IDR:
412 	case NVME_DSMGMT_IDW:
413 	default:
414 		/* Not supported yet */
415 		nvmet_req_complete(req, 0);
416 		return;
417 	}
418 }
419 
420 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
421 {
422 	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
423 	struct bio *bio = NULL;
424 	sector_t sector;
425 	sector_t nr_sector;
426 	int ret;
427 
428 	if (!nvmet_check_transfer_len(req, 0))
429 		return;
430 
431 	sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
432 	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
433 		(req->ns->blksize_shift - 9));
434 
435 	ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
436 			GFP_KERNEL, &bio, 0);
437 	if (bio) {
438 		bio->bi_private = req;
439 		bio->bi_end_io = nvmet_bio_done;
440 		submit_bio(bio);
441 	} else {
442 		nvmet_req_complete(req, errno_to_nvme_status(req, ret));
443 	}
444 }
445 
446 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
447 {
448 	switch (req->cmd->common.opcode) {
449 	case nvme_cmd_read:
450 	case nvme_cmd_write:
451 		req->execute = nvmet_bdev_execute_rw;
452 		if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
453 			req->metadata_len = nvmet_rw_metadata_len(req);
454 		return 0;
455 	case nvme_cmd_flush:
456 		req->execute = nvmet_bdev_execute_flush;
457 		return 0;
458 	case nvme_cmd_dsm:
459 		req->execute = nvmet_bdev_execute_dsm;
460 		return 0;
461 	case nvme_cmd_write_zeroes:
462 		req->execute = nvmet_bdev_execute_write_zeroes;
463 		return 0;
464 	default:
465 		return nvmet_report_invalid_opcode(req);
466 	}
467 }
468