xref: /spdk/lib/nvmf/ctrlr_bdev.c (revision 60982c759db49b4f4579f16e3b24df0725ba4b94)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (C) 2017 Intel Corporation. All rights reserved.
3  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
4  *   Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5  */
6 
7 #include "spdk/stdinc.h"
8 
9 #include "nvmf_internal.h"
10 
11 #include "spdk/bdev.h"
12 #include "spdk/endian.h"
13 #include "spdk/thread.h"
14 #include "spdk/likely.h"
15 #include "spdk/nvme.h"
16 #include "spdk/nvmf_cmd.h"
17 #include "spdk/nvmf_spec.h"
18 #include "spdk/trace.h"
19 #include "spdk/scsi_spec.h"
20 #include "spdk/string.h"
21 #include "spdk/util.h"
22 
23 #include "spdk/log.h"
24 
25 static bool
26 nvmf_subsystem_bdev_io_type_supported(struct spdk_nvmf_subsystem *subsystem,
27 				      enum spdk_bdev_io_type io_type)
28 {
29 	struct spdk_nvmf_ns *ns;
30 
31 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
32 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
33 		if (ns->bdev == NULL) {
34 			continue;
35 		}
36 
37 		if (!spdk_bdev_io_type_supported(ns->bdev, io_type)) {
38 			SPDK_DEBUGLOG(nvmf,
39 				      "Subsystem %s namespace %u (%s) does not support io_type %d\n",
40 				      spdk_nvmf_subsystem_get_nqn(subsystem),
41 				      ns->opts.nsid, spdk_bdev_get_name(ns->bdev), (int)io_type);
42 			return false;
43 		}
44 	}
45 
46 	SPDK_DEBUGLOG(nvmf, "All devices in Subsystem %s support io_type %d\n",
47 		      spdk_nvmf_subsystem_get_nqn(subsystem), (int)io_type);
48 	return true;
49 }
50 
51 bool
52 nvmf_ctrlr_dsm_supported(struct spdk_nvmf_ctrlr *ctrlr)
53 {
54 	return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_UNMAP);
55 }
56 
57 bool
58 nvmf_ctrlr_write_zeroes_supported(struct spdk_nvmf_ctrlr *ctrlr)
59 {
60 	return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_WRITE_ZEROES);
61 }
62 
63 bool
64 nvmf_ctrlr_copy_supported(struct spdk_nvmf_ctrlr *ctrlr)
65 {
66 	return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_COPY);
67 }
68 
69 static void
70 nvmf_bdev_ctrlr_complete_cmd(struct spdk_bdev_io *bdev_io, bool success,
71 			     void *cb_arg)
72 {
73 	struct spdk_nvmf_request	*req = cb_arg;
74 	struct spdk_nvme_cpl		*response = &req->rsp->nvme_cpl;
75 	int				first_sc = 0, first_sct = 0, sc = 0, sct = 0;
76 	uint32_t			cdw0 = 0;
77 	struct spdk_nvmf_request	*first_req = req->first_fused_req;
78 
79 	if (spdk_unlikely(first_req != NULL)) {
80 		/* fused commands - get status for both operations */
81 		struct spdk_nvme_cpl *first_response = &first_req->rsp->nvme_cpl;
82 
83 		spdk_bdev_io_get_nvme_fused_status(bdev_io, &cdw0, &first_sct, &first_sc, &sct, &sc);
84 		first_response->cdw0 = cdw0;
85 		first_response->status.sc = first_sc;
86 		first_response->status.sct = first_sct;
87 
88 		/* first request should be completed */
89 		spdk_nvmf_request_complete(first_req);
90 		req->first_fused_req = NULL;
91 	} else {
92 		spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc);
93 	}
94 
95 	response->cdw0 = cdw0;
96 	response->status.sc = sc;
97 	response->status.sct = sct;
98 
99 	spdk_nvmf_request_complete(req);
100 	spdk_bdev_free_io(bdev_io);
101 }
102 
103 static void
104 nvmf_bdev_ctrlr_complete_admin_cmd(struct spdk_bdev_io *bdev_io, bool success,
105 				   void *cb_arg)
106 {
107 	struct spdk_nvmf_request *req = cb_arg;
108 
109 	if (req->cmd_cb_fn) {
110 		req->cmd_cb_fn(req);
111 	}
112 
113 	nvmf_bdev_ctrlr_complete_cmd(bdev_io, success, req);
114 }
115 
116 void
117 nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *nsdata,
118 			    bool dif_insert_or_strip)
119 {
120 	struct spdk_bdev *bdev = ns->bdev;
121 	uint64_t num_blocks;
122 	uint32_t phys_blocklen;
123 	uint32_t max_copy;
124 
125 	num_blocks = spdk_bdev_get_num_blocks(bdev);
126 
127 	nsdata->nsze = num_blocks;
128 	nsdata->ncap = num_blocks;
129 	nsdata->nuse = num_blocks;
130 	nsdata->nlbaf = 0;
131 	nsdata->flbas.format = 0;
132 	nsdata->flbas.msb_format = 0;
133 	nsdata->nacwu = spdk_bdev_get_acwu(bdev) - 1; /* nacwu is 0-based */
134 	if (!dif_insert_or_strip) {
135 		nsdata->lbaf[0].ms = spdk_bdev_get_md_size(bdev);
136 		nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_block_size(bdev));
137 		if (nsdata->lbaf[0].ms != 0) {
138 			nsdata->flbas.extended = 1;
139 			nsdata->mc.extended = 1;
140 			nsdata->mc.pointer = 0;
141 			nsdata->dps.md_start = spdk_bdev_is_dif_head_of_md(bdev);
142 			/* NVMf library doesn't process PRACT and PRCHK flags, we
143 			 * leave the use of extended LBA buffer to users.
144 			 */
145 			nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_DISABLE;
146 		}
147 	} else {
148 		nsdata->lbaf[0].ms = 0;
149 		nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_data_block_size(bdev));
150 	}
151 
152 	phys_blocklen = spdk_bdev_get_physical_block_size(bdev);
153 	assert(phys_blocklen > 0);
154 	/* Linux driver uses min(nawupf, npwg) to set physical_block_size */
155 	nsdata->nsfeat.optperf = 1;
156 	nsdata->nsfeat.ns_atomic_write_unit = 1;
157 	nsdata->npwg = (phys_blocklen >> nsdata->lbaf[0].lbads) - 1;
158 	nsdata->nawupf = nsdata->npwg;
159 	nsdata->npwa = nsdata->npwg;
160 	nsdata->npdg = nsdata->npwg;
161 	nsdata->npda = nsdata->npwg;
162 
163 	if (spdk_bdev_get_write_unit_size(bdev) == 1) {
164 		nsdata->noiob = spdk_bdev_get_optimal_io_boundary(bdev);
165 	}
166 	nsdata->nmic.can_share = 1;
167 	if (ns->ptpl_file != NULL) {
168 		nsdata->nsrescap.rescap.persist = 1;
169 	}
170 	nsdata->nsrescap.rescap.write_exclusive = 1;
171 	nsdata->nsrescap.rescap.exclusive_access = 1;
172 	nsdata->nsrescap.rescap.write_exclusive_reg_only = 1;
173 	nsdata->nsrescap.rescap.exclusive_access_reg_only = 1;
174 	nsdata->nsrescap.rescap.write_exclusive_all_reg = 1;
175 	nsdata->nsrescap.rescap.exclusive_access_all_reg = 1;
176 	nsdata->nsrescap.rescap.ignore_existing_key = 1;
177 
178 	SPDK_STATIC_ASSERT(sizeof(nsdata->nguid) == sizeof(ns->opts.nguid), "size mismatch");
179 	memcpy(nsdata->nguid, ns->opts.nguid, sizeof(nsdata->nguid));
180 
181 	SPDK_STATIC_ASSERT(sizeof(nsdata->eui64) == sizeof(ns->opts.eui64), "size mismatch");
182 	memcpy(&nsdata->eui64, ns->opts.eui64, sizeof(nsdata->eui64));
183 
184 	/* For now we support just one source range for copy command */
185 	nsdata->msrc = 0;
186 
187 	max_copy = spdk_bdev_get_max_copy(bdev);
188 	if (max_copy == 0 || max_copy > UINT16_MAX) {
189 		/* Zero means copy size is unlimited */
190 		nsdata->mcl = UINT16_MAX;
191 		nsdata->mssrl = UINT16_MAX;
192 	} else {
193 		nsdata->mcl = max_copy;
194 		nsdata->mssrl = max_copy;
195 	}
196 }
197 
198 static void
199 nvmf_bdev_ctrlr_get_rw_params(const struct spdk_nvme_cmd *cmd, uint64_t *start_lba,
200 			      uint64_t *num_blocks)
201 {
202 	/* SLBA: CDW10 and CDW11 */
203 	*start_lba = from_le64(&cmd->cdw10);
204 
205 	/* NLB: CDW12 bits 15:00, 0's based */
206 	*num_blocks = (from_le32(&cmd->cdw12) & 0xFFFFu) + 1;
207 }
208 
209 static bool
210 nvmf_bdev_ctrlr_lba_in_range(uint64_t bdev_num_blocks, uint64_t io_start_lba,
211 			     uint64_t io_num_blocks)
212 {
213 	if (io_start_lba + io_num_blocks > bdev_num_blocks ||
214 	    io_start_lba + io_num_blocks < io_start_lba) {
215 		return false;
216 	}
217 
218 	return true;
219 }
220 
221 static void
222 nvmf_ctrlr_process_io_cmd_resubmit(void *arg)
223 {
224 	struct spdk_nvmf_request *req = arg;
225 	int rc;
226 
227 	rc = nvmf_ctrlr_process_io_cmd(req);
228 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
229 		spdk_nvmf_request_complete(req);
230 	}
231 }
232 
233 static void
234 nvmf_ctrlr_process_admin_cmd_resubmit(void *arg)
235 {
236 	struct spdk_nvmf_request *req = arg;
237 	int rc;
238 
239 	rc = nvmf_ctrlr_process_admin_cmd(req);
240 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
241 		spdk_nvmf_request_complete(req);
242 	}
243 }
244 
245 static void
246 nvmf_bdev_ctrl_queue_io(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
247 			struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn, void *cb_arg)
248 {
249 	int rc;
250 
251 	req->bdev_io_wait.bdev = bdev;
252 	req->bdev_io_wait.cb_fn = cb_fn;
253 	req->bdev_io_wait.cb_arg = cb_arg;
254 
255 	rc = spdk_bdev_queue_io_wait(bdev, ch, &req->bdev_io_wait);
256 	if (rc != 0) {
257 		assert(false);
258 	}
259 	req->qpair->group->stat.pending_bdev_io++;
260 }
261 
262 bool
263 nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev)
264 {
265 	return spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY);
266 }
267 
268 int
269 nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
270 			 struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
271 {
272 	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);
273 	uint32_t block_size = spdk_bdev_get_block_size(bdev);
274 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
275 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
276 	uint64_t start_lba;
277 	uint64_t num_blocks;
278 	int rc;
279 
280 	nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks);
281 
282 	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) {
283 		SPDK_ERRLOG("end of media\n");
284 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
285 		rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE;
286 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
287 	}
288 
289 	if (spdk_unlikely(num_blocks * block_size > req->length)) {
290 		SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n",
291 			    num_blocks, block_size, req->length);
292 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
293 		rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
294 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
295 	}
296 
297 	assert(!spdk_nvmf_request_using_zcopy(req));
298 
299 	rc = spdk_bdev_readv_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks,
300 				    nvmf_bdev_ctrlr_complete_cmd, req);
301 	if (spdk_unlikely(rc)) {
302 		if (rc == -ENOMEM) {
303 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
304 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
305 		}
306 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
307 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
308 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
309 	}
310 
311 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
312 }
313 
314 int
315 nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
316 			  struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
317 {
318 	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);
319 	uint32_t block_size = spdk_bdev_get_block_size(bdev);
320 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
321 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
322 	uint64_t start_lba;
323 	uint64_t num_blocks;
324 	int rc;
325 
326 	nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks);
327 
328 	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) {
329 		SPDK_ERRLOG("end of media\n");
330 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
331 		rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE;
332 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
333 	}
334 
335 	if (spdk_unlikely(num_blocks * block_size > req->length)) {
336 		SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n",
337 			    num_blocks, block_size, req->length);
338 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
339 		rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
340 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
341 	}
342 
343 	assert(!spdk_nvmf_request_using_zcopy(req));
344 
345 	rc = spdk_bdev_writev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks,
346 				     nvmf_bdev_ctrlr_complete_cmd, req);
347 	if (spdk_unlikely(rc)) {
348 		if (rc == -ENOMEM) {
349 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
350 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
351 		}
352 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
353 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
354 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
355 	}
356 
357 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
358 }
359 
360 int
361 nvmf_bdev_ctrlr_compare_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
362 			    struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
363 {
364 	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);
365 	uint32_t block_size = spdk_bdev_get_block_size(bdev);
366 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
367 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
368 	uint64_t start_lba;
369 	uint64_t num_blocks;
370 	int rc;
371 
372 	nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks);
373 
374 	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) {
375 		SPDK_ERRLOG("end of media\n");
376 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
377 		rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE;
378 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
379 	}
380 
381 	if (spdk_unlikely(num_blocks * block_size > req->length)) {
382 		SPDK_ERRLOG("Compare NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n",
383 			    num_blocks, block_size, req->length);
384 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
385 		rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
386 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
387 	}
388 
389 	rc = spdk_bdev_comparev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks,
390 				       nvmf_bdev_ctrlr_complete_cmd, req);
391 	if (spdk_unlikely(rc)) {
392 		if (rc == -ENOMEM) {
393 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
394 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
395 		}
396 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
397 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
398 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
399 	}
400 
401 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
402 }
403 
404 int
405 nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
406 				      struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req)
407 {
408 	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);
409 	uint32_t block_size = spdk_bdev_get_block_size(bdev);
410 	struct spdk_nvme_cmd *cmp_cmd = &cmp_req->cmd->nvme_cmd;
411 	struct spdk_nvme_cmd *write_cmd = &write_req->cmd->nvme_cmd;
412 	struct spdk_nvme_cpl *rsp = &write_req->rsp->nvme_cpl;
413 	uint64_t write_start_lba, cmp_start_lba;
414 	uint64_t write_num_blocks, cmp_num_blocks;
415 	int rc;
416 
417 	nvmf_bdev_ctrlr_get_rw_params(cmp_cmd, &cmp_start_lba, &cmp_num_blocks);
418 	nvmf_bdev_ctrlr_get_rw_params(write_cmd, &write_start_lba, &write_num_blocks);
419 
420 	if (spdk_unlikely(write_start_lba != cmp_start_lba || write_num_blocks != cmp_num_blocks)) {
421 		SPDK_ERRLOG("Fused command start lba / num blocks mismatch\n");
422 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
423 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
424 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
425 	}
426 
427 	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, write_start_lba,
428 			  write_num_blocks))) {
429 		SPDK_ERRLOG("end of media\n");
430 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
431 		rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE;
432 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
433 	}
434 
435 	if (spdk_unlikely(write_num_blocks * block_size > write_req->length)) {
436 		SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n",
437 			    write_num_blocks, block_size, write_req->length);
438 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
439 		rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
440 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
441 	}
442 
443 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ch, cmp_req->iov, cmp_req->iovcnt, write_req->iov,
444 			write_req->iovcnt, write_start_lba, write_num_blocks, nvmf_bdev_ctrlr_complete_cmd, write_req);
445 	if (spdk_unlikely(rc)) {
446 		if (rc == -ENOMEM) {
447 			nvmf_bdev_ctrl_queue_io(cmp_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, cmp_req);
448 			nvmf_bdev_ctrl_queue_io(write_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, write_req);
449 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
450 		}
451 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
452 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
453 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
454 	}
455 
456 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
457 }
458 
459 int
460 nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
461 				 struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
462 {
463 	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);
464 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
465 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
466 	uint64_t start_lba;
467 	uint64_t num_blocks;
468 	int rc;
469 
470 	nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks);
471 
472 	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) {
473 		SPDK_ERRLOG("end of media\n");
474 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
475 		rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE;
476 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
477 	}
478 
479 	rc = spdk_bdev_write_zeroes_blocks(desc, ch, start_lba, num_blocks,
480 					   nvmf_bdev_ctrlr_complete_cmd, req);
481 	if (spdk_unlikely(rc)) {
482 		if (rc == -ENOMEM) {
483 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
484 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
485 		}
486 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
487 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
488 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
489 	}
490 
491 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
492 }
493 
494 int
495 nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
496 			  struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
497 {
498 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
499 	int rc;
500 
501 	/* As for NVMeoF controller, SPDK always set volatile write
502 	 * cache bit to 1, return success for those block devices
503 	 * which can't support FLUSH command.
504 	 */
505 	if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH)) {
506 		response->status.sct = SPDK_NVME_SCT_GENERIC;
507 		response->status.sc = SPDK_NVME_SC_SUCCESS;
508 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
509 	}
510 
511 	rc = spdk_bdev_flush_blocks(desc, ch, 0, spdk_bdev_get_num_blocks(bdev),
512 				    nvmf_bdev_ctrlr_complete_cmd, req);
513 	if (spdk_unlikely(rc)) {
514 		if (rc == -ENOMEM) {
515 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
516 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
517 		}
518 		response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
519 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
520 	}
521 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
522 }
523 
524 struct nvmf_bdev_ctrlr_unmap {
525 	struct spdk_nvmf_request	*req;
526 	uint32_t			count;
527 	struct spdk_bdev_desc		*desc;
528 	struct spdk_bdev		*bdev;
529 	struct spdk_io_channel		*ch;
530 	uint32_t			range_index;
531 };
532 
533 static void
534 nvmf_bdev_ctrlr_unmap_cpl(struct spdk_bdev_io *bdev_io, bool success,
535 			  void *cb_arg)
536 {
537 	struct nvmf_bdev_ctrlr_unmap *unmap_ctx = cb_arg;
538 	struct spdk_nvmf_request	*req = unmap_ctx->req;
539 	struct spdk_nvme_cpl		*response = &req->rsp->nvme_cpl;
540 	int				sc, sct;
541 	uint32_t			cdw0;
542 
543 	unmap_ctx->count--;
544 
545 	if (response->status.sct == SPDK_NVME_SCT_GENERIC &&
546 	    response->status.sc == SPDK_NVME_SC_SUCCESS) {
547 		spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc);
548 		response->cdw0 = cdw0;
549 		response->status.sc = sc;
550 		response->status.sct = sct;
551 	}
552 
553 	if (unmap_ctx->count == 0) {
554 		spdk_nvmf_request_complete(req);
555 		free(unmap_ctx);
556 	}
557 	spdk_bdev_free_io(bdev_io);
558 }
559 
560 static int nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
561 				 struct spdk_io_channel *ch, struct spdk_nvmf_request *req,
562 				 struct nvmf_bdev_ctrlr_unmap *unmap_ctx);
563 static void
564 nvmf_bdev_ctrlr_unmap_resubmit(void *arg)
565 {
566 	struct nvmf_bdev_ctrlr_unmap *unmap_ctx = arg;
567 	struct spdk_nvmf_request *req = unmap_ctx->req;
568 	struct spdk_bdev_desc *desc = unmap_ctx->desc;
569 	struct spdk_bdev *bdev = unmap_ctx->bdev;
570 	struct spdk_io_channel *ch = unmap_ctx->ch;
571 
572 	nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, unmap_ctx);
573 }
574 
575 static int
576 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
577 		      struct spdk_io_channel *ch, struct spdk_nvmf_request *req,
578 		      struct nvmf_bdev_ctrlr_unmap *unmap_ctx)
579 {
580 	uint16_t nr, i;
581 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
582 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
583 	struct spdk_iov_xfer ix;
584 	uint64_t lba;
585 	uint32_t lba_count;
586 	int rc;
587 
588 	nr = cmd->cdw10_bits.dsm.nr + 1;
589 	if (nr * sizeof(struct spdk_nvme_dsm_range) > req->length) {
590 		SPDK_ERRLOG("Dataset Management number of ranges > SGL length\n");
591 		response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
592 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
593 	}
594 
595 	if (unmap_ctx == NULL) {
596 		unmap_ctx = calloc(1, sizeof(*unmap_ctx));
597 		if (!unmap_ctx) {
598 			response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
599 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
600 		}
601 
602 		unmap_ctx->req = req;
603 		unmap_ctx->desc = desc;
604 		unmap_ctx->ch = ch;
605 		unmap_ctx->bdev = bdev;
606 
607 		response->status.sct = SPDK_NVME_SCT_GENERIC;
608 		response->status.sc = SPDK_NVME_SC_SUCCESS;
609 	} else {
610 		unmap_ctx->count--;	/* dequeued */
611 	}
612 
613 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
614 
615 	for (i = unmap_ctx->range_index; i < nr; i++) {
616 		struct spdk_nvme_dsm_range dsm_range = { 0 };
617 
618 		spdk_iov_xfer_to_buf(&ix, &dsm_range, sizeof(dsm_range));
619 
620 		lba = dsm_range.starting_lba;
621 		lba_count = dsm_range.length;
622 
623 		unmap_ctx->count++;
624 
625 		rc = spdk_bdev_unmap_blocks(desc, ch, lba, lba_count,
626 					    nvmf_bdev_ctrlr_unmap_cpl, unmap_ctx);
627 		if (rc) {
628 			if (rc == -ENOMEM) {
629 				nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_bdev_ctrlr_unmap_resubmit, unmap_ctx);
630 				/* Unmap was not yet submitted to bdev */
631 				/* unmap_ctx->count will be decremented when the request is dequeued */
632 				return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
633 			}
634 			response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
635 			unmap_ctx->count--;
636 			/* We can't return here - we may have to wait for any other
637 				* unmaps already sent to complete */
638 			break;
639 		}
640 		unmap_ctx->range_index++;
641 	}
642 
643 	if (unmap_ctx->count == 0) {
644 		free(unmap_ctx);
645 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
646 	}
647 
648 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
649 }
650 
651 int
652 nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
653 			struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
654 {
655 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
656 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
657 
658 	if (cmd->cdw11_bits.dsm.ad) {
659 		return nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, NULL);
660 	}
661 
662 	response->status.sct = SPDK_NVME_SCT_GENERIC;
663 	response->status.sc = SPDK_NVME_SC_SUCCESS;
664 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
665 }
666 
667 int
668 nvmf_bdev_ctrlr_copy_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
669 			 struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
670 {
671 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
672 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
673 	uint64_t sdlba = ((uint64_t)cmd->cdw11 << 32) + cmd->cdw10;
674 	struct spdk_nvme_scc_source_range range = { 0 };
675 	struct spdk_iov_xfer ix;
676 	int rc;
677 
678 	SPDK_DEBUGLOG(nvmf, "Copy command: SDLBA %lu, NR %u, desc format %u, PRINFOR %u, "
679 		      "DTYPE %u, STCW %u, PRINFOW %u, FUA %u, LR %u\n",
680 		      sdlba,
681 		      cmd->cdw12_bits.copy.nr,
682 		      cmd->cdw12_bits.copy.df,
683 		      cmd->cdw12_bits.copy.prinfor,
684 		      cmd->cdw12_bits.copy.dtype,
685 		      cmd->cdw12_bits.copy.stcw,
686 		      cmd->cdw12_bits.copy.prinfow,
687 		      cmd->cdw12_bits.copy.fua,
688 		      cmd->cdw12_bits.copy.lr);
689 
690 	if (spdk_unlikely(req->length != (cmd->cdw12_bits.copy.nr + 1) *
691 			  sizeof(struct spdk_nvme_scc_source_range))) {
692 		response->status.sct = SPDK_NVME_SCT_GENERIC;
693 		response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
694 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
695 	}
696 
697 	/*
698 	 * We support only one source range, and rely on this with the xfer
699 	 * below.
700 	 */
701 	if (cmd->cdw12_bits.copy.nr > 0) {
702 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
703 		response->status.sc = SPDK_NVME_SC_CMD_SIZE_LIMIT_SIZE_EXCEEDED;
704 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
705 	}
706 
707 	if (cmd->cdw12_bits.copy.df != 0) {
708 		response->status.sct = SPDK_NVME_SCT_GENERIC;
709 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
710 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
711 	}
712 
713 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
714 	spdk_iov_xfer_to_buf(&ix, &range, sizeof(range));
715 
716 	rc = spdk_bdev_copy_blocks(desc, ch, sdlba, range.slba, range.nlb + 1,
717 				   nvmf_bdev_ctrlr_complete_cmd, req);
718 	if (spdk_unlikely(rc)) {
719 		if (rc == -ENOMEM) {
720 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
721 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
722 		}
723 
724 		response->status.sct = SPDK_NVME_SCT_GENERIC;
725 		response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
726 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
727 	}
728 
729 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
730 }
731 
732 int
733 nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
734 				 struct spdk_io_channel *ch, struct spdk_nvmf_request *req)
735 {
736 	int rc;
737 
738 	if (spdk_unlikely(req->iovcnt > 1)) {
739 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
740 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
741 		req->rsp->nvme_cpl.status.dnr = 1;
742 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
743 	}
744 
745 	rc = spdk_bdev_nvme_io_passthru(desc, ch, &req->cmd->nvme_cmd, req->iov[0].iov_base, req->length,
746 					nvmf_bdev_ctrlr_complete_cmd, req);
747 	if (spdk_unlikely(rc)) {
748 		if (rc == -ENOMEM) {
749 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
750 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
751 		}
752 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
753 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
754 		req->rsp->nvme_cpl.status.dnr = 1;
755 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
756 	}
757 
758 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
759 }
760 
761 int
762 spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
763 		struct spdk_io_channel *ch, struct spdk_nvmf_request *req,
764 		spdk_nvmf_nvme_passthru_cmd_cb cb_fn)
765 {
766 	int rc;
767 
768 	if (spdk_unlikely(req->iovcnt > 1)) {
769 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
770 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
771 		req->rsp->nvme_cpl.status.dnr = 1;
772 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
773 	}
774 
775 	req->cmd_cb_fn = cb_fn;
776 
777 	rc = spdk_bdev_nvme_admin_passthru(desc, ch, &req->cmd->nvme_cmd, req->iov[0].iov_base, req->length,
778 					   nvmf_bdev_ctrlr_complete_admin_cmd, req);
779 	if (spdk_unlikely(rc)) {
780 		if (rc == -ENOMEM) {
781 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req);
782 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
783 		}
784 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
785 		if (rc == -ENOTSUP) {
786 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
787 		} else {
788 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
789 		}
790 
791 		req->rsp->nvme_cpl.status.dnr = 1;
792 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
793 	}
794 
795 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
796 }
797 
798 static void
799 nvmf_bdev_ctrlr_complete_abort_cmd(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
800 {
801 	struct spdk_nvmf_request *req = cb_arg;
802 
803 	if (success) {
804 		req->rsp->nvme_cpl.cdw0 &= ~1U;
805 	}
806 
807 	spdk_nvmf_request_complete(req);
808 	spdk_bdev_free_io(bdev_io);
809 }
810 
811 int
812 spdk_nvmf_bdev_ctrlr_abort_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
813 			       struct spdk_io_channel *ch, struct spdk_nvmf_request *req,
814 			       struct spdk_nvmf_request *req_to_abort)
815 {
816 	int rc;
817 
818 	assert((req->rsp->nvme_cpl.cdw0 & 1U) != 0);
819 
820 	rc = spdk_bdev_abort(desc, ch, req_to_abort, nvmf_bdev_ctrlr_complete_abort_cmd, req);
821 	if (spdk_likely(rc == 0)) {
822 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
823 	} else if (rc == -ENOMEM) {
824 		nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req);
825 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
826 	} else {
827 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
828 	}
829 }
830 
831 bool
832 nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev *bdev, struct spdk_nvme_cmd *cmd,
833 			    struct spdk_dif_ctx *dif_ctx)
834 {
835 	uint32_t init_ref_tag, dif_check_flags = 0;
836 	int rc;
837 	struct spdk_dif_ctx_init_ext_opts dif_opts;
838 
839 	if (spdk_bdev_get_md_size(bdev) == 0) {
840 		return false;
841 	}
842 
843 	/* Initial Reference Tag is the lower 32 bits of the start LBA. */
844 	init_ref_tag = (uint32_t)from_le64(&cmd->cdw10);
845 
846 	if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_REFTAG)) {
847 		dif_check_flags |= SPDK_DIF_FLAGS_REFTAG_CHECK;
848 	}
849 
850 	if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_GUARD)) {
851 		dif_check_flags |= SPDK_DIF_FLAGS_GUARD_CHECK;
852 	}
853 
854 	dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
855 	dif_opts.dif_pi_format = SPDK_DIF_PI_FORMAT_16;
856 	rc = spdk_dif_ctx_init(dif_ctx,
857 			       spdk_bdev_get_block_size(bdev),
858 			       spdk_bdev_get_md_size(bdev),
859 			       spdk_bdev_is_md_interleaved(bdev),
860 			       spdk_bdev_is_dif_head_of_md(bdev),
861 			       spdk_bdev_get_dif_type(bdev),
862 			       dif_check_flags,
863 			       init_ref_tag, 0, 0, 0, 0, &dif_opts);
864 
865 	return (rc == 0) ? true : false;
866 }
867 
868 static void
869 nvmf_bdev_ctrlr_zcopy_start_complete(struct spdk_bdev_io *bdev_io, bool success,
870 				     void *cb_arg)
871 {
872 	struct spdk_nvmf_request	*req = cb_arg;
873 	struct iovec *iov;
874 	int iovcnt = 0;
875 
876 	if (spdk_unlikely(!success)) {
877 		int                     sc = 0, sct = 0;
878 		uint32_t                cdw0 = 0;
879 		struct spdk_nvme_cpl    *response = &req->rsp->nvme_cpl;
880 		spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc);
881 
882 		response->cdw0 = cdw0;
883 		response->status.sc = sc;
884 		response->status.sct = sct;
885 
886 		spdk_bdev_free_io(bdev_io);
887 		spdk_nvmf_request_complete(req);
888 		return;
889 	}
890 
891 	spdk_bdev_io_get_iovec(bdev_io, &iov, &iovcnt);
892 
893 	assert(iovcnt <= NVMF_REQ_MAX_BUFFERS);
894 	assert(iovcnt > 0);
895 
896 	req->iovcnt = iovcnt;
897 
898 	assert(req->iov == iov);
899 
900 	/* backward compatible */
901 	req->data = req->iov[0].iov_base;
902 
903 	req->zcopy_bdev_io = bdev_io; /* Preserve the bdev_io for the end zcopy */
904 
905 	spdk_nvmf_request_complete(req);
906 	/* Don't free the bdev_io here as it is needed for the END ZCOPY */
907 }
908 
909 int
910 nvmf_bdev_ctrlr_zcopy_start(struct spdk_bdev *bdev,
911 			    struct spdk_bdev_desc *desc,
912 			    struct spdk_io_channel *ch,
913 			    struct spdk_nvmf_request *req)
914 {
915 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
916 	uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev);
917 	uint32_t block_size = spdk_bdev_get_block_size(bdev);
918 	uint64_t start_lba;
919 	uint64_t num_blocks;
920 	int rc;
921 
922 	nvmf_bdev_ctrlr_get_rw_params(&req->cmd->nvme_cmd, &start_lba, &num_blocks);
923 
924 	if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) {
925 		SPDK_ERRLOG("end of media\n");
926 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
927 		rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE;
928 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
929 	}
930 
931 	if (spdk_unlikely(num_blocks * block_size > req->length)) {
932 		SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n",
933 			    num_blocks, block_size, req->length);
934 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
935 		rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID;
936 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
937 	}
938 
939 	bool populate = (req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_READ) ? true : false;
940 
941 	rc = spdk_bdev_zcopy_start(desc, ch, req->iov, req->iovcnt, start_lba,
942 				   num_blocks, populate, nvmf_bdev_ctrlr_zcopy_start_complete, req);
943 	if (spdk_unlikely(rc != 0)) {
944 		if (rc == -ENOMEM) {
945 			nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req);
946 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
947 		}
948 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
949 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
950 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
951 	}
952 
953 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
954 }
955 
956 static void
957 nvmf_bdev_ctrlr_zcopy_end_complete(struct spdk_bdev_io *bdev_io, bool success,
958 				   void *cb_arg)
959 {
960 	struct spdk_nvmf_request	*req = cb_arg;
961 
962 	if (spdk_unlikely(!success)) {
963 		int                     sc = 0, sct = 0;
964 		uint32_t                cdw0 = 0;
965 		struct spdk_nvme_cpl    *response = &req->rsp->nvme_cpl;
966 		spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc);
967 
968 		response->cdw0 = cdw0;
969 		response->status.sc = sc;
970 		response->status.sct = sct;
971 	}
972 
973 	spdk_bdev_free_io(bdev_io);
974 	req->zcopy_bdev_io = NULL;
975 	spdk_nvmf_request_complete(req);
976 }
977 
978 void
979 nvmf_bdev_ctrlr_zcopy_end(struct spdk_nvmf_request *req, bool commit)
980 {
981 	int rc __attribute__((unused));
982 
983 	rc = spdk_bdev_zcopy_end(req->zcopy_bdev_io, commit, nvmf_bdev_ctrlr_zcopy_end_complete, req);
984 
985 	/* The only way spdk_bdev_zcopy_end() can fail is if we pass a bdev_io type that isn't ZCOPY */
986 	assert(rc == 0);
987 }
988