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/module.h>
10 #include "nvmet.h"
11 
nvmet_bdev_set_limits(struct block_device * bdev,struct nvme_id_ns * id)12 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)
13 {
14 	const struct queue_limits *ql = &bdev_get_queue(bdev)->limits;
15 	/* Number of logical blocks per physical block. */
16 	const u32 lpp = ql->physical_block_size / ql->logical_block_size;
17 	/* Logical blocks per physical block, 0's based. */
18 	const __le16 lpp0b = to0based(lpp);
19 
20 	/*
21 	 * For NVMe 1.2 and later, bit 1 indicates that the fields NAWUN,
22 	 * NAWUPF, and NACWU are defined for this namespace and should be
23 	 * used by the host for this namespace instead of the AWUN, AWUPF,
24 	 * and ACWU fields in the Identify Controller data structure. If
25 	 * any of these fields are zero that means that the corresponding
26 	 * field from the identify controller data structure should be used.
27 	 */
28 	id->nsfeat |= 1 << 1;
29 	id->nawun = lpp0b;
30 	id->nawupf = lpp0b;
31 	id->nacwu = lpp0b;
32 
33 	/*
34 	 * Bit 4 indicates that the fields NPWG, NPWA, NPDG, NPDA, and
35 	 * NOWS are defined for this namespace and should be used by
36 	 * the host for I/O optimization.
37 	 */
38 	id->nsfeat |= 1 << 4;
39 	/* NPWG = Namespace Preferred Write Granularity. 0's based */
40 	id->npwg = lpp0b;
41 	/* NPWA = Namespace Preferred Write Alignment. 0's based */
42 	id->npwa = id->npwg;
43 	/* NPDG = Namespace Preferred Deallocate Granularity. 0's based */
44 	id->npdg = to0based(ql->discard_granularity / ql->logical_block_size);
45 	/* NPDG = Namespace Preferred Deallocate Alignment */
46 	id->npda = id->npdg;
47 	/* NOWS = Namespace Optimal Write Size */
48 	id->nows = to0based(ql->io_opt / ql->logical_block_size);
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) {
54 		blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
55 		ns->bdev = NULL;
56 	}
57 }
58 
nvmet_bdev_ns_enable_integrity(struct nvmet_ns * ns)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 
nvmet_bdev_ns_enable(struct nvmet_ns * ns)75 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
76 {
77 	int ret;
78 
79 	ns->bdev = blkdev_get_by_path(ns->device_path,
80 			FMODE_READ | FMODE_WRITE, NULL);
81 	if (IS_ERR(ns->bdev)) {
82 		ret = PTR_ERR(ns->bdev);
83 		if (ret != -ENOTBLK) {
84 			pr_err("failed to open block device %s: (%ld)\n",
85 					ns->device_path, PTR_ERR(ns->bdev));
86 		}
87 		ns->bdev = NULL;
88 		return ret;
89 	}
90 	ns->size = bdev_nr_bytes(ns->bdev);
91 	ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
92 
93 	ns->pi_type = 0;
94 	ns->metadata_size = 0;
95 	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY_T10))
96 		nvmet_bdev_ns_enable_integrity(ns);
97 
98 	if (bdev_is_zoned(ns->bdev)) {
99 		if (!nvmet_bdev_zns_enable(ns)) {
100 			nvmet_bdev_ns_disable(ns);
101 			return -EINVAL;
102 		}
103 		ns->csi = NVME_CSI_ZNS;
104 	}
105 
106 	return 0;
107 }
108 
nvmet_bdev_ns_revalidate(struct nvmet_ns * ns)109 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
110 {
111 	ns->size = bdev_nr_bytes(ns->bdev);
112 }
113 
blk_to_nvme_status(struct nvmet_req * req,blk_status_t blk_sts)114 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
115 {
116 	u16 status = NVME_SC_SUCCESS;
117 
118 	if (likely(blk_sts == BLK_STS_OK))
119 		return status;
120 	/*
121 	 * Right now there exists M : 1 mapping between block layer error
122 	 * to the NVMe status code (see nvme_error_status()). For consistency,
123 	 * when we reverse map we use most appropriate NVMe Status code from
124 	 * the group of the NVMe staus codes used in the nvme_error_status().
125 	 */
126 	switch (blk_sts) {
127 	case BLK_STS_NOSPC:
128 		status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
129 		req->error_loc = offsetof(struct nvme_rw_command, length);
130 		break;
131 	case BLK_STS_TARGET:
132 		status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
133 		req->error_loc = offsetof(struct nvme_rw_command, slba);
134 		break;
135 	case BLK_STS_NOTSUPP:
136 		req->error_loc = offsetof(struct nvme_common_command, opcode);
137 		switch (req->cmd->common.opcode) {
138 		case nvme_cmd_dsm:
139 		case nvme_cmd_write_zeroes:
140 			status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
141 			break;
142 		default:
143 			status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
144 		}
145 		break;
146 	case BLK_STS_MEDIUM:
147 		status = NVME_SC_ACCESS_DENIED;
148 		req->error_loc = offsetof(struct nvme_rw_command, nsid);
149 		break;
150 	case BLK_STS_IOERR:
151 	default:
152 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
153 		req->error_loc = offsetof(struct nvme_common_command, opcode);
154 	}
155 
156 	switch (req->cmd->common.opcode) {
157 	case nvme_cmd_read:
158 	case nvme_cmd_write:
159 		req->error_slba = le64_to_cpu(req->cmd->rw.slba);
160 		break;
161 	case nvme_cmd_write_zeroes:
162 		req->error_slba =
163 			le64_to_cpu(req->cmd->write_zeroes.slba);
164 		break;
165 	default:
166 		req->error_slba = 0;
167 	}
168 	return status;
169 }
170 
nvmet_bio_done(struct bio * bio)171 static void nvmet_bio_done(struct bio *bio)
172 {
173 	struct nvmet_req *req = bio->bi_private;
174 
175 	nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
176 	nvmet_req_bio_put(req, bio);
177 }
178 
179 #ifdef CONFIG_BLK_DEV_INTEGRITY
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)180 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
181 				struct sg_mapping_iter *miter)
182 {
183 	struct blk_integrity *bi;
184 	struct bio_integrity_payload *bip;
185 	int rc;
186 	size_t resid, len;
187 
188 	bi = bdev_get_integrity(req->ns->bdev);
189 	if (unlikely(!bi)) {
190 		pr_err("Unable to locate bio_integrity\n");
191 		return -ENODEV;
192 	}
193 
194 	bip = bio_integrity_alloc(bio, GFP_NOIO,
195 					bio_max_segs(req->metadata_sg_cnt));
196 	if (IS_ERR(bip)) {
197 		pr_err("Unable to allocate bio_integrity_payload\n");
198 		return PTR_ERR(bip);
199 	}
200 
201 	bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
202 	/* virtual start sector must be in integrity interval units */
203 	bip_set_seed(bip, bio->bi_iter.bi_sector >>
204 		     (bi->interval_exp - SECTOR_SHIFT));
205 
206 	resid = bip->bip_iter.bi_size;
207 	while (resid > 0 && sg_miter_next(miter)) {
208 		len = min_t(size_t, miter->length, resid);
209 		rc = bio_integrity_add_page(bio, miter->page, len,
210 					    offset_in_page(miter->addr));
211 		if (unlikely(rc != len)) {
212 			pr_err("bio_integrity_add_page() failed; %d\n", rc);
213 			sg_miter_stop(miter);
214 			return -ENOMEM;
215 		}
216 
217 		resid -= len;
218 		if (len < miter->length)
219 			miter->consumed -= miter->length - len;
220 	}
221 	sg_miter_stop(miter);
222 
223 	return 0;
224 }
225 #else
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)226 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
227 				struct sg_mapping_iter *miter)
228 {
229 	return -EINVAL;
230 }
231 #endif /* CONFIG_BLK_DEV_INTEGRITY */
232 
nvmet_bdev_execute_rw(struct nvmet_req * req)233 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
234 {
235 	unsigned int sg_cnt = req->sg_cnt;
236 	struct bio *bio;
237 	struct scatterlist *sg;
238 	struct blk_plug plug;
239 	sector_t sector;
240 	int op, i, rc;
241 	struct sg_mapping_iter prot_miter;
242 	unsigned int iter_flags;
243 	unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
244 
245 	if (!nvmet_check_transfer_len(req, total_len))
246 		return;
247 
248 	if (!req->sg_cnt) {
249 		nvmet_req_complete(req, 0);
250 		return;
251 	}
252 
253 	if (req->cmd->rw.opcode == nvme_cmd_write) {
254 		op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
255 		if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
256 			op |= REQ_FUA;
257 		iter_flags = SG_MITER_TO_SG;
258 	} else {
259 		op = REQ_OP_READ;
260 		iter_flags = SG_MITER_FROM_SG;
261 	}
262 
263 	if (is_pci_p2pdma_page(sg_page(req->sg)))
264 		op |= REQ_NOMERGE;
265 
266 	sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
267 
268 	if (nvmet_use_inline_bvec(req)) {
269 		bio = &req->b.inline_bio;
270 		bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
271 	} else {
272 		bio = bio_alloc(GFP_KERNEL, bio_max_segs(sg_cnt));
273 	}
274 	bio_set_dev(bio, req->ns->bdev);
275 	bio->bi_iter.bi_sector = sector;
276 	bio->bi_private = req;
277 	bio->bi_end_io = nvmet_bio_done;
278 	bio->bi_opf = op;
279 
280 	blk_start_plug(&plug);
281 	if (req->metadata_len)
282 		sg_miter_start(&prot_miter, req->metadata_sg,
283 			       req->metadata_sg_cnt, iter_flags);
284 
285 	for_each_sg(req->sg, sg, req->sg_cnt, i) {
286 		while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
287 				!= sg->length) {
288 			struct bio *prev = bio;
289 
290 			if (req->metadata_len) {
291 				rc = nvmet_bdev_alloc_bip(req, bio,
292 							  &prot_miter);
293 				if (unlikely(rc)) {
294 					bio_io_error(bio);
295 					return;
296 				}
297 			}
298 
299 			bio = bio_alloc(GFP_KERNEL, bio_max_segs(sg_cnt));
300 			bio_set_dev(bio, req->ns->bdev);
301 			bio->bi_iter.bi_sector = sector;
302 			bio->bi_opf = op;
303 
304 			bio_chain(bio, prev);
305 			submit_bio(prev);
306 		}
307 
308 		sector += sg->length >> 9;
309 		sg_cnt--;
310 	}
311 
312 	if (req->metadata_len) {
313 		rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
314 		if (unlikely(rc)) {
315 			bio_io_error(bio);
316 			return;
317 		}
318 	}
319 
320 	submit_bio(bio);
321 	blk_finish_plug(&plug);
322 }
323 
nvmet_bdev_execute_flush(struct nvmet_req * req)324 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
325 {
326 	struct bio *bio = &req->b.inline_bio;
327 
328 	if (!nvmet_check_transfer_len(req, 0))
329 		return;
330 
331 	bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
332 	bio_set_dev(bio, req->ns->bdev);
333 	bio->bi_private = req;
334 	bio->bi_end_io = nvmet_bio_done;
335 	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
336 
337 	submit_bio(bio);
338 }
339 
nvmet_bdev_flush(struct nvmet_req * req)340 u16 nvmet_bdev_flush(struct nvmet_req *req)
341 {
342 	if (blkdev_issue_flush(req->ns->bdev))
343 		return NVME_SC_INTERNAL | NVME_SC_DNR;
344 	return 0;
345 }
346 
nvmet_bdev_discard_range(struct nvmet_req * req,struct nvme_dsm_range * range,struct bio ** bio)347 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
348 		struct nvme_dsm_range *range, struct bio **bio)
349 {
350 	struct nvmet_ns *ns = req->ns;
351 	int ret;
352 
353 	ret = __blkdev_issue_discard(ns->bdev,
354 			nvmet_lba_to_sect(ns, range->slba),
355 			le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
356 			GFP_KERNEL, 0, bio);
357 	if (ret && ret != -EOPNOTSUPP) {
358 		req->error_slba = le64_to_cpu(range->slba);
359 		return errno_to_nvme_status(req, ret);
360 	}
361 	return NVME_SC_SUCCESS;
362 }
363 
nvmet_bdev_execute_discard(struct nvmet_req * req)364 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
365 {
366 	struct nvme_dsm_range range;
367 	struct bio *bio = NULL;
368 	int i;
369 	u16 status;
370 
371 	for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
372 		status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
373 				sizeof(range));
374 		if (status)
375 			break;
376 
377 		status = nvmet_bdev_discard_range(req, &range, &bio);
378 		if (status)
379 			break;
380 	}
381 
382 	if (bio) {
383 		bio->bi_private = req;
384 		bio->bi_end_io = nvmet_bio_done;
385 		if (status)
386 			bio_io_error(bio);
387 		else
388 			submit_bio(bio);
389 	} else {
390 		nvmet_req_complete(req, status);
391 	}
392 }
393 
nvmet_bdev_execute_dsm(struct nvmet_req * req)394 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
395 {
396 	if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
397 		return;
398 
399 	switch (le32_to_cpu(req->cmd->dsm.attributes)) {
400 	case NVME_DSMGMT_AD:
401 		nvmet_bdev_execute_discard(req);
402 		return;
403 	case NVME_DSMGMT_IDR:
404 	case NVME_DSMGMT_IDW:
405 	default:
406 		/* Not supported yet */
407 		nvmet_req_complete(req, 0);
408 		return;
409 	}
410 }
411 
nvmet_bdev_execute_write_zeroes(struct nvmet_req * req)412 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
413 {
414 	struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
415 	struct bio *bio = NULL;
416 	sector_t sector;
417 	sector_t nr_sector;
418 	int ret;
419 
420 	if (!nvmet_check_transfer_len(req, 0))
421 		return;
422 
423 	sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
424 	nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
425 		(req->ns->blksize_shift - 9));
426 
427 	ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
428 			GFP_KERNEL, &bio, 0);
429 	if (bio) {
430 		bio->bi_private = req;
431 		bio->bi_end_io = nvmet_bio_done;
432 		submit_bio(bio);
433 	} else {
434 		nvmet_req_complete(req, errno_to_nvme_status(req, ret));
435 	}
436 }
437 
nvmet_bdev_parse_io_cmd(struct nvmet_req * req)438 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
439 {
440 	switch (req->cmd->common.opcode) {
441 	case nvme_cmd_read:
442 	case nvme_cmd_write:
443 		req->execute = nvmet_bdev_execute_rw;
444 		if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
445 			req->metadata_len = nvmet_rw_metadata_len(req);
446 		return 0;
447 	case nvme_cmd_flush:
448 		req->execute = nvmet_bdev_execute_flush;
449 		return 0;
450 	case nvme_cmd_dsm:
451 		req->execute = nvmet_bdev_execute_dsm;
452 		return 0;
453 	case nvme_cmd_write_zeroes:
454 		req->execute = nvmet_bdev_execute_write_zeroes;
455 		return 0;
456 	default:
457 		return nvmet_report_invalid_opcode(req);
458 	}
459 }
460