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/module.h>
9 #include "nvmet.h"
10
nvmet_bdev_set_limits(struct block_device * bdev,struct nvme_id_ns * id)11 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)
12 {
13 const struct queue_limits *ql = &bdev_get_queue(bdev)->limits;
14 /* Number of logical blocks per physical block. */
15 const u32 lpp = ql->physical_block_size / ql->logical_block_size;
16 /* Logical blocks per physical block, 0's based. */
17 const __le16 lpp0b = to0based(lpp);
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(ql->discard_granularity / ql->logical_block_size);
44 /* NPDG = Namespace Preferred Deallocate Alignment */
45 id->npda = id->npdg;
46 /* NOWS = Namespace Optimal Write Size */
47 id->nows = to0based(ql->io_opt / ql->logical_block_size);
48 }
49
nvmet_bdev_ns_disable(struct nvmet_ns * ns)50 void nvmet_bdev_ns_disable(struct nvmet_ns *ns)
51 {
52 if (ns->bdev) {
53 blkdev_put(ns->bdev, FMODE_WRITE | FMODE_READ);
54 ns->bdev = NULL;
55 }
56 }
57
nvmet_bdev_ns_enable_integrity(struct nvmet_ns * ns)58 static void nvmet_bdev_ns_enable_integrity(struct nvmet_ns *ns)
59 {
60 struct blk_integrity *bi = bdev_get_integrity(ns->bdev);
61
62 if (bi) {
63 ns->metadata_size = bi->tuple_size;
64 if (bi->profile == &t10_pi_type1_crc)
65 ns->pi_type = NVME_NS_DPS_PI_TYPE1;
66 else if (bi->profile == &t10_pi_type3_crc)
67 ns->pi_type = NVME_NS_DPS_PI_TYPE3;
68 else
69 /* Unsupported metadata type */
70 ns->metadata_size = 0;
71 }
72 }
73
nvmet_bdev_ns_enable(struct nvmet_ns * ns)74 int nvmet_bdev_ns_enable(struct nvmet_ns *ns)
75 {
76 int ret;
77
78 ns->bdev = blkdev_get_by_path(ns->device_path,
79 FMODE_READ | FMODE_WRITE, NULL);
80 if (IS_ERR(ns->bdev)) {
81 ret = PTR_ERR(ns->bdev);
82 if (ret != -ENOTBLK) {
83 pr_err("failed to open block device %s: (%ld)\n",
84 ns->device_path, PTR_ERR(ns->bdev));
85 }
86 ns->bdev = NULL;
87 return ret;
88 }
89 ns->size = i_size_read(ns->bdev->bd_inode);
90 ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
91
92 ns->pi_type = 0;
93 ns->metadata_size = 0;
94 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY_T10))
95 nvmet_bdev_ns_enable_integrity(ns);
96
97 if (bdev_is_zoned(ns->bdev)) {
98 if (!nvmet_bdev_zns_enable(ns)) {
99 nvmet_bdev_ns_disable(ns);
100 return -EINVAL;
101 }
102 ns->csi = NVME_CSI_ZNS;
103 }
104
105 return 0;
106 }
107
nvmet_bdev_ns_revalidate(struct nvmet_ns * ns)108 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns)
109 {
110 ns->size = i_size_read(ns->bdev->bd_inode);
111 }
112
blk_to_nvme_status(struct nvmet_req * req,blk_status_t blk_sts)113 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts)
114 {
115 u16 status = NVME_SC_SUCCESS;
116
117 if (likely(blk_sts == BLK_STS_OK))
118 return status;
119 /*
120 * Right now there exists M : 1 mapping between block layer error
121 * to the NVMe status code (see nvme_error_status()). For consistency,
122 * when we reverse map we use most appropriate NVMe Status code from
123 * the group of the NVMe staus codes used in the nvme_error_status().
124 */
125 switch (blk_sts) {
126 case BLK_STS_NOSPC:
127 status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
128 req->error_loc = offsetof(struct nvme_rw_command, length);
129 break;
130 case BLK_STS_TARGET:
131 status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
132 req->error_loc = offsetof(struct nvme_rw_command, slba);
133 break;
134 case BLK_STS_NOTSUPP:
135 req->error_loc = offsetof(struct nvme_common_command, opcode);
136 switch (req->cmd->common.opcode) {
137 case nvme_cmd_dsm:
138 case nvme_cmd_write_zeroes:
139 status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
140 break;
141 default:
142 status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
143 }
144 break;
145 case BLK_STS_MEDIUM:
146 status = NVME_SC_ACCESS_DENIED;
147 req->error_loc = offsetof(struct nvme_rw_command, nsid);
148 break;
149 case BLK_STS_IOERR:
150 default:
151 status = NVME_SC_INTERNAL | NVME_SC_DNR;
152 req->error_loc = offsetof(struct nvme_common_command, opcode);
153 }
154
155 switch (req->cmd->common.opcode) {
156 case nvme_cmd_read:
157 case nvme_cmd_write:
158 req->error_slba = le64_to_cpu(req->cmd->rw.slba);
159 break;
160 case nvme_cmd_write_zeroes:
161 req->error_slba =
162 le64_to_cpu(req->cmd->write_zeroes.slba);
163 break;
164 default:
165 req->error_slba = 0;
166 }
167 return status;
168 }
169
nvmet_bio_done(struct bio * bio)170 static void nvmet_bio_done(struct bio *bio)
171 {
172 struct nvmet_req *req = bio->bi_private;
173
174 nvmet_req_complete(req, blk_to_nvme_status(req, bio->bi_status));
175 nvmet_req_bio_put(req, bio);
176 }
177
178 #ifdef CONFIG_BLK_DEV_INTEGRITY
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)179 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
180 struct sg_mapping_iter *miter)
181 {
182 struct blk_integrity *bi;
183 struct bio_integrity_payload *bip;
184 int rc;
185 size_t resid, len;
186
187 bi = bdev_get_integrity(req->ns->bdev);
188 if (unlikely(!bi)) {
189 pr_err("Unable to locate bio_integrity\n");
190 return -ENODEV;
191 }
192
193 bip = bio_integrity_alloc(bio, GFP_NOIO,
194 bio_max_segs(req->metadata_sg_cnt));
195 if (IS_ERR(bip)) {
196 pr_err("Unable to allocate bio_integrity_payload\n");
197 return PTR_ERR(bip);
198 }
199
200 bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
201 /* virtual start sector must be in integrity interval units */
202 bip_set_seed(bip, bio->bi_iter.bi_sector >>
203 (bi->interval_exp - SECTOR_SHIFT));
204
205 resid = bip->bip_iter.bi_size;
206 while (resid > 0 && sg_miter_next(miter)) {
207 len = min_t(size_t, miter->length, resid);
208 rc = bio_integrity_add_page(bio, miter->page, len,
209 offset_in_page(miter->addr));
210 if (unlikely(rc != len)) {
211 pr_err("bio_integrity_add_page() failed; %d\n", rc);
212 sg_miter_stop(miter);
213 return -ENOMEM;
214 }
215
216 resid -= len;
217 if (len < miter->length)
218 miter->consumed -= miter->length - len;
219 }
220 sg_miter_stop(miter);
221
222 return 0;
223 }
224 #else
nvmet_bdev_alloc_bip(struct nvmet_req * req,struct bio * bio,struct sg_mapping_iter * miter)225 static int nvmet_bdev_alloc_bip(struct nvmet_req *req, struct bio *bio,
226 struct sg_mapping_iter *miter)
227 {
228 return -EINVAL;
229 }
230 #endif /* CONFIG_BLK_DEV_INTEGRITY */
231
nvmet_bdev_execute_rw(struct nvmet_req * req)232 static void nvmet_bdev_execute_rw(struct nvmet_req *req)
233 {
234 unsigned int sg_cnt = req->sg_cnt;
235 struct bio *bio;
236 struct scatterlist *sg;
237 struct blk_plug plug;
238 sector_t sector;
239 int op, i, rc;
240 struct sg_mapping_iter prot_miter;
241 unsigned int iter_flags;
242 unsigned int total_len = nvmet_rw_data_len(req) + req->metadata_len;
243
244 if (!nvmet_check_transfer_len(req, total_len))
245 return;
246
247 if (!req->sg_cnt) {
248 nvmet_req_complete(req, 0);
249 return;
250 }
251
252 if (req->cmd->rw.opcode == nvme_cmd_write) {
253 op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
254 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
255 op |= REQ_FUA;
256 iter_flags = SG_MITER_TO_SG;
257 } else {
258 op = REQ_OP_READ;
259 iter_flags = SG_MITER_FROM_SG;
260 }
261
262 if (is_pci_p2pdma_page(sg_page(req->sg)))
263 op |= REQ_NOMERGE;
264
265 sector = nvmet_lba_to_sect(req->ns, req->cmd->rw.slba);
266
267 if (nvmet_use_inline_bvec(req)) {
268 bio = &req->b.inline_bio;
269 bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
270 } else {
271 bio = bio_alloc(GFP_KERNEL, bio_max_segs(sg_cnt));
272 }
273 bio_set_dev(bio, req->ns->bdev);
274 bio->bi_iter.bi_sector = sector;
275 bio->bi_private = req;
276 bio->bi_end_io = nvmet_bio_done;
277 bio->bi_opf = op;
278
279 blk_start_plug(&plug);
280 if (req->metadata_len)
281 sg_miter_start(&prot_miter, req->metadata_sg,
282 req->metadata_sg_cnt, iter_flags);
283
284 for_each_sg(req->sg, sg, req->sg_cnt, i) {
285 while (bio_add_page(bio, sg_page(sg), sg->length, sg->offset)
286 != sg->length) {
287 struct bio *prev = bio;
288
289 if (req->metadata_len) {
290 rc = nvmet_bdev_alloc_bip(req, bio,
291 &prot_miter);
292 if (unlikely(rc)) {
293 bio_io_error(bio);
294 return;
295 }
296 }
297
298 bio = bio_alloc(GFP_KERNEL, bio_max_segs(sg_cnt));
299 bio_set_dev(bio, req->ns->bdev);
300 bio->bi_iter.bi_sector = sector;
301 bio->bi_opf = op;
302
303 bio_chain(bio, prev);
304 submit_bio(prev);
305 }
306
307 sector += sg->length >> 9;
308 sg_cnt--;
309 }
310
311 if (req->metadata_len) {
312 rc = nvmet_bdev_alloc_bip(req, bio, &prot_miter);
313 if (unlikely(rc)) {
314 bio_io_error(bio);
315 return;
316 }
317 }
318
319 submit_bio(bio);
320 blk_finish_plug(&plug);
321 }
322
nvmet_bdev_execute_flush(struct nvmet_req * req)323 static void nvmet_bdev_execute_flush(struct nvmet_req *req)
324 {
325 struct bio *bio = &req->b.inline_bio;
326
327 if (!nvmet_check_transfer_len(req, 0))
328 return;
329
330 bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
331 bio_set_dev(bio, req->ns->bdev);
332 bio->bi_private = req;
333 bio->bi_end_io = nvmet_bio_done;
334 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
335
336 submit_bio(bio);
337 }
338
nvmet_bdev_flush(struct nvmet_req * req)339 u16 nvmet_bdev_flush(struct nvmet_req *req)
340 {
341 if (blkdev_issue_flush(req->ns->bdev))
342 return NVME_SC_INTERNAL | NVME_SC_DNR;
343 return 0;
344 }
345
nvmet_bdev_discard_range(struct nvmet_req * req,struct nvme_dsm_range * range,struct bio ** bio)346 static u16 nvmet_bdev_discard_range(struct nvmet_req *req,
347 struct nvme_dsm_range *range, struct bio **bio)
348 {
349 struct nvmet_ns *ns = req->ns;
350 int ret;
351
352 ret = __blkdev_issue_discard(ns->bdev,
353 nvmet_lba_to_sect(ns, range->slba),
354 le32_to_cpu(range->nlb) << (ns->blksize_shift - 9),
355 GFP_KERNEL, 0, bio);
356 if (ret && ret != -EOPNOTSUPP) {
357 req->error_slba = le64_to_cpu(range->slba);
358 return errno_to_nvme_status(req, ret);
359 }
360 return NVME_SC_SUCCESS;
361 }
362
nvmet_bdev_execute_discard(struct nvmet_req * req)363 static void nvmet_bdev_execute_discard(struct nvmet_req *req)
364 {
365 struct nvme_dsm_range range;
366 struct bio *bio = NULL;
367 int i;
368 u16 status;
369
370 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
371 status = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
372 sizeof(range));
373 if (status)
374 break;
375
376 status = nvmet_bdev_discard_range(req, &range, &bio);
377 if (status)
378 break;
379 }
380
381 if (bio) {
382 bio->bi_private = req;
383 bio->bi_end_io = nvmet_bio_done;
384 if (status)
385 bio_io_error(bio);
386 else
387 submit_bio(bio);
388 } else {
389 nvmet_req_complete(req, status);
390 }
391 }
392
nvmet_bdev_execute_dsm(struct nvmet_req * req)393 static void nvmet_bdev_execute_dsm(struct nvmet_req *req)
394 {
395 if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
396 return;
397
398 switch (le32_to_cpu(req->cmd->dsm.attributes)) {
399 case NVME_DSMGMT_AD:
400 nvmet_bdev_execute_discard(req);
401 return;
402 case NVME_DSMGMT_IDR:
403 case NVME_DSMGMT_IDW:
404 default:
405 /* Not supported yet */
406 nvmet_req_complete(req, 0);
407 return;
408 }
409 }
410
nvmet_bdev_execute_write_zeroes(struct nvmet_req * req)411 static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)
412 {
413 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
414 struct bio *bio = NULL;
415 sector_t sector;
416 sector_t nr_sector;
417 int ret;
418
419 if (!nvmet_check_transfer_len(req, 0))
420 return;
421
422 sector = nvmet_lba_to_sect(req->ns, write_zeroes->slba);
423 nr_sector = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
424 (req->ns->blksize_shift - 9));
425
426 ret = __blkdev_issue_zeroout(req->ns->bdev, sector, nr_sector,
427 GFP_KERNEL, &bio, 0);
428 if (bio) {
429 bio->bi_private = req;
430 bio->bi_end_io = nvmet_bio_done;
431 submit_bio(bio);
432 } else {
433 nvmet_req_complete(req, errno_to_nvme_status(req, ret));
434 }
435 }
436
nvmet_bdev_parse_io_cmd(struct nvmet_req * req)437 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)
438 {
439 switch (req->cmd->common.opcode) {
440 case nvme_cmd_read:
441 case nvme_cmd_write:
442 req->execute = nvmet_bdev_execute_rw;
443 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns))
444 req->metadata_len = nvmet_rw_metadata_len(req);
445 return 0;
446 case nvme_cmd_flush:
447 req->execute = nvmet_bdev_execute_flush;
448 return 0;
449 case nvme_cmd_dsm:
450 req->execute = nvmet_bdev_execute_dsm;
451 return 0;
452 case nvme_cmd_write_zeroes:
453 req->execute = nvmet_bdev_execute_write_zeroes;
454 return 0;
455 default:
456 return nvmet_report_invalid_opcode(req);
457 }
458 }
459