projects / fio.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
engines/io_uring: fix leak of 'ld' in error path
[fio.git] / engines / nvme.c
CommitLineData
ca59c41c 1// SPDX-License-Identifier: GPL-2.0
b3d5e3fd
AK		 2/*
3 * nvme structure declarations and helper functions for the
4 * io_uring_cmd engine.
5 */
6
7#include "nvme.h"
5163f35e 8#include "../crc/crc-t10dif.h"
08371767 9#include "../crc/crc64.h"
b3d5e3fd 10
4885a6eb
VF		 11static inline __u64 get_slba(struct nvme_data *data, struct io_u *io_u)
12{
13 if (data->lba_ext)
14 return io_u->offset / data->lba_ext;
15 else
16 return io_u->offset >> data->lba_shift;
17}
18
19static inline __u32 get_nlb(struct nvme_data *data, struct io_u *io_u)
20{
21 if (data->lba_ext)
22 return io_u->xfer_buflen / data->lba_ext - 1;
23 else
24 return (io_u->xfer_buflen >> data->lba_shift) - 1;
25}
26
5163f35e
AK		 27static void fio_nvme_generate_pi_16b_guard(struct nvme_data *data,
28 struct io_u *io_u,
29 struct nvme_cmd_ext_io_opts *opts)
30{
31 struct nvme_pi_data *pi_data = io_u->engine_data;
32 struct nvme_16b_guard_pif *pi;
33 unsigned char *buf = io_u->xfer_buf;
34 unsigned char *md_buf = io_u->mmap_data;
35 __u64 slba = get_slba(data, io_u);
36 __u32 nlb = get_nlb(data, io_u) + 1;
37 __u32 lba_num = 0;
38 __u16 guard = 0;
39
40 if (data->pi_loc) {
41 if (data->lba_ext)
42 pi_data->interval = data->lba_ext - data->ms;
43 else
44 pi_data->interval = 0;
45 } else {
46 if (data->lba_ext)
47 pi_data->interval = data->lba_ext - sizeof(struct nvme_16b_guard_pif);
48 else
49 pi_data->interval = data->ms - sizeof(struct nvme_16b_guard_pif);
50 }
51
52 if (io_u->ddir != DDIR_WRITE)
53 return;
54
55 while (lba_num < nlb) {
56 if (data->lba_ext)
57 pi = (struct nvme_16b_guard_pif *)(buf + pi_data->interval);
58 else
59 pi = (struct nvme_16b_guard_pif *)(md_buf + pi_data->interval);
60
61 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
62 if (data->lba_ext) {
63 guard = fio_crc_t10dif(0, buf, pi_data->interval);
64 } else {
65 guard = fio_crc_t10dif(0, buf, data->lba_size);
66 guard = fio_crc_t10dif(guard, md_buf, pi_data->interval);
67 }
68 pi->guard = cpu_to_be16(guard);
69 }
70
71 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_APP)
72 pi->apptag = cpu_to_be16(pi_data->apptag);
73
74 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
75 switch (data->pi_type) {
76 case NVME_NS_DPS_PI_TYPE1:
77 case NVME_NS_DPS_PI_TYPE2:
78 pi->srtag = cpu_to_be32((__u32)slba + lba_num);
79 break;
80 case NVME_NS_DPS_PI_TYPE3:
81 break;
82 }
83 }
84 if (data->lba_ext) {
85 buf += data->lba_ext;
86 } else {
87 buf += data->lba_size;
88 md_buf += data->ms;
89 }
90 lba_num++;
91 }
92}
93
94static int fio_nvme_verify_pi_16b_guard(struct nvme_data *data,
95 struct io_u *io_u)
96{
97 struct nvme_pi_data *pi_data = io_u->engine_data;
98 struct nvme_16b_guard_pif *pi;
99 struct fio_file *f = io_u->file;
100 unsigned char *buf = io_u->xfer_buf;
101 unsigned char *md_buf = io_u->mmap_data;
102 __u64 slba = get_slba(data, io_u);
103 __u32 nlb = get_nlb(data, io_u) + 1;
104 __u32 lba_num = 0;
105 __u16 unmask_app, unmask_app_exp, guard = 0;
106
107 while (lba_num < nlb) {
108 if (data->lba_ext)
109 pi = (struct nvme_16b_guard_pif *)(buf + pi_data->interval);
110 else
111 pi = (struct nvme_16b_guard_pif *)(md_buf + pi_data->interval);
112
113 if (data->pi_type == NVME_NS_DPS_PI_TYPE3) {
114 if (pi->apptag == NVME_PI_APP_DISABLE &&
115 pi->srtag == NVME_PI_REF_DISABLE)
116 goto next;
117 } else if (data->pi_type == NVME_NS_DPS_PI_TYPE1 ||
118 data->pi_type == NVME_NS_DPS_PI_TYPE2) {
119 if (pi->apptag == NVME_PI_APP_DISABLE)
120 goto next;
121 }
122
123 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
124 if (data->lba_ext) {
125 guard = fio_crc_t10dif(0, buf, pi_data->interval);
126 } else {
127 guard = fio_crc_t10dif(0, buf, data->lba_size);
128 guard = fio_crc_t10dif(guard, md_buf, pi_data->interval);
129 }
130 if (be16_to_cpu(pi->guard) != guard) {
131 log_err("%s: Guard compare error: LBA: %llu Expected=%x, Actual=%x\n",
132 f->file_name, (unsigned long long)slba,
133 guard, be16_to_cpu(pi->guard));
134 return -EIO;
135 }
136 }
137
138 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_APP) {
139 unmask_app = be16_to_cpu(pi->apptag) & pi_data->apptag_mask;
140 unmask_app_exp = pi_data->apptag & pi_data->apptag_mask;
141 if (unmask_app != unmask_app_exp) {
142 log_err("%s: APPTAG compare error: LBA: %llu Expected=%x, Actual=%x\n",
143 f->file_name, (unsigned long long)slba,
144 unmask_app_exp, unmask_app);
145 return -EIO;
146 }
147 }
148
149 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
150 switch (data->pi_type) {
151 case NVME_NS_DPS_PI_TYPE1:
152 case NVME_NS_DPS_PI_TYPE2:
153 if (be32_to_cpu(pi->srtag) !=
154 ((__u32)slba + lba_num)) {
155 log_err("%s: REFTAG compare error: LBA: %llu Expected=%x, Actual=%x\n",
156 f->file_name, (unsigned long long)slba,
157 (__u32)slba + lba_num,
158 be32_to_cpu(pi->srtag));
159 return -EIO;
160 }
161 break;
162 case NVME_NS_DPS_PI_TYPE3:
163 break;
164 }
165 }
166next:
167 if (data->lba_ext) {
168 buf += data->lba_ext;
169 } else {
170 buf += data->lba_size;
171 md_buf += data->ms;
172 }
173 lba_num++;
174 }
175
176 return 0;
177}
178
08371767
AK		 179static void fio_nvme_generate_pi_64b_guard(struct nvme_data *data,
180 struct io_u *io_u,
181 struct nvme_cmd_ext_io_opts *opts)
182{
183 struct nvme_pi_data *pi_data = io_u->engine_data;
184 struct nvme_64b_guard_pif *pi;
185 unsigned char *buf = io_u->xfer_buf;
186 unsigned char *md_buf = io_u->mmap_data;
187 uint64_t guard = 0;
188 __u64 slba = get_slba(data, io_u);
189 __u32 nlb = get_nlb(data, io_u) + 1;
190 __u32 lba_num = 0;
191
192 if (data->pi_loc) {
193 if (data->lba_ext)
194 pi_data->interval = data->lba_ext - data->ms;
195 else
196 pi_data->interval = 0;
197 } else {
198 if (data->lba_ext)
199 pi_data->interval = data->lba_ext - sizeof(struct nvme_64b_guard_pif);
200 else
201 pi_data->interval = data->ms - sizeof(struct nvme_64b_guard_pif);
202 }
203
204 if (io_u->ddir != DDIR_WRITE)
205 return;
206
207 while (lba_num < nlb) {
208 if (data->lba_ext)
209 pi = (struct nvme_64b_guard_pif *)(buf + pi_data->interval);
210 else
211 pi = (struct nvme_64b_guard_pif *)(md_buf + pi_data->interval);
212
213 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
214 if (data->lba_ext) {
215 guard = fio_crc64_nvme(0, buf, pi_data->interval);
216 } else {
217 guard = fio_crc64_nvme(0, buf, data->lba_size);
218 guard = fio_crc64_nvme(guard, md_buf, pi_data->interval);
219 }
220 pi->guard = cpu_to_be64(guard);
221 }
222
223 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_APP)
224 pi->apptag = cpu_to_be16(pi_data->apptag);
225
226 if (opts->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
227 switch (data->pi_type) {
228 case NVME_NS_DPS_PI_TYPE1:
229 case NVME_NS_DPS_PI_TYPE2:
230 put_unaligned_be48(slba + lba_num, pi->srtag);
231 break;
232 case NVME_NS_DPS_PI_TYPE3:
233 break;
234 }
235 }
236 if (data->lba_ext) {
237 buf += data->lba_ext;
238 } else {
239 buf += data->lba_size;
240 md_buf += data->ms;
241 }
242 lba_num++;
243 }
244}
245
246static int fio_nvme_verify_pi_64b_guard(struct nvme_data *data,
247 struct io_u *io_u)
248{
249 struct nvme_pi_data *pi_data = io_u->engine_data;
250 struct nvme_64b_guard_pif *pi;
251 struct fio_file *f = io_u->file;
252 unsigned char *buf = io_u->xfer_buf;
253 unsigned char *md_buf = io_u->mmap_data;
254 __u64 slba = get_slba(data, io_u);
255 __u64 ref, ref_exp, guard = 0;
256 __u32 nlb = get_nlb(data, io_u) + 1;
257 __u32 lba_num = 0;
258 __u16 unmask_app, unmask_app_exp;
259
260 while (lba_num < nlb) {
261 if (data->lba_ext)
262 pi = (struct nvme_64b_guard_pif *)(buf + pi_data->interval);
263 else
264 pi = (struct nvme_64b_guard_pif *)(md_buf + pi_data->interval);
265
266 if (data->pi_type == NVME_NS_DPS_PI_TYPE3) {
267 if (pi->apptag == NVME_PI_APP_DISABLE &&
268 fio_nvme_pi_ref_escape(pi->srtag))
269 goto next;
270 } else if (data->pi_type == NVME_NS_DPS_PI_TYPE1 ||
271 data->pi_type == NVME_NS_DPS_PI_TYPE2) {
272 if (pi->apptag == NVME_PI_APP_DISABLE)
273 goto next;
274 }
275
276 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_GUARD) {
277 if (data->lba_ext) {
278 guard = fio_crc64_nvme(0, buf, pi_data->interval);
279 } else {
280 guard = fio_crc64_nvme(0, buf, data->lba_size);
281 guard = fio_crc64_nvme(guard, md_buf, pi_data->interval);
282 }
283 if (be64_to_cpu((uint64_t)pi->guard) != guard) {
284 log_err("%s: Guard compare error: LBA: %llu Expected=%llx, Actual=%llx\n",
285 f->file_name, (unsigned long long)slba,
286 guard, be64_to_cpu((uint64_t)pi->guard));
287 return -EIO;
288 }
289 }
290
291 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_APP) {
292 unmask_app = be16_to_cpu(pi->apptag) & pi_data->apptag_mask;
293 unmask_app_exp = pi_data->apptag & pi_data->apptag_mask;
294 if (unmask_app != unmask_app_exp) {
295 log_err("%s: APPTAG compare error: LBA: %llu Expected=%x, Actual=%x\n",
296 f->file_name, (unsigned long long)slba,
297 unmask_app_exp, unmask_app);
298 return -EIO;
299 }
300 }
301
302 if (pi_data->io_flags & NVME_IO_PRINFO_PRCHK_REF) {
303 switch (data->pi_type) {
304 case NVME_NS_DPS_PI_TYPE1:
305 case NVME_NS_DPS_PI_TYPE2:
306 ref = get_unaligned_be48(pi->srtag);
307 ref_exp = (slba + lba_num) & ((1ULL << 48) - 1);
308 if (ref != ref_exp) {
309 log_err("%s: REFTAG compare error: LBA: %llu Expected=%llx, Actual=%llx\n",
310 f->file_name, (unsigned long long)slba,
311 ref_exp, ref);
312 return -EIO;
313 }
314 break;
315 case NVME_NS_DPS_PI_TYPE3:
316 break;
317 }
318 }
319next:
320 if (data->lba_ext) {
321 buf += data->lba_ext;
322 } else {
323 buf += data->lba_size;
324 md_buf += data->ms;
325 }
326 lba_num++;
327 }
328
329 return 0;
330}
4885a6eb
VF		 331void fio_nvme_uring_cmd_trim_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
332 struct nvme_dsm_range *dsm)
333{
334 struct nvme_data *data = FILE_ENG_DATA(io_u->file);
335
336 cmd->opcode = nvme_cmd_dsm;
337 cmd->nsid = data->nsid;
338 cmd->cdw10 = 0;
339 cmd->cdw11 = NVME_ATTRIBUTE_DEALLOCATE;
340 cmd->addr = (__u64) (uintptr_t) dsm;
341 cmd->data_len = sizeof(*dsm);
342
343 dsm->slba = get_slba(data, io_u);
344 /* nlb is a 1-based value for deallocate */
345 dsm->nlb = get_nlb(data, io_u) + 1;
346}
347
b3d5e3fd 348int fio_nvme_uring_cmd_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
4885a6eb 349 struct iovec *iov, struct nvme_dsm_range *dsm)
b3d5e3fd
AK		 350{
351 struct nvme_data *data = FILE_ENG_DATA(io_u->file);
352 __u64 slba;
353 __u32 nlb;
354
355 memset(cmd, 0, sizeof(struct nvme_uring_cmd));
356
4885a6eb
VF		 357 switch (io_u->ddir) {
358 case DDIR_READ:
b3d5e3fd 359 cmd->opcode = nvme_cmd_read;
4885a6eb
VF		 360 break;
361 case DDIR_WRITE:
b3d5e3fd 362 cmd->opcode = nvme_cmd_write;
4885a6eb
VF		 363 break;
364 case DDIR_TRIM:
365 fio_nvme_uring_cmd_trim_prep(cmd, io_u, dsm);
366 return 0;
367 default:
b3d5e3fd 368 return -ENOTSUP;
345fa8fd 369 }
b3d5e3fd 370
4885a6eb
VF		 371 slba = get_slba(data, io_u);
372 nlb = get_nlb(data, io_u);
373
b3d5e3fd
AK		 374 /* cdw10 and cdw11 represent starting lba */
375 cmd->cdw10 = slba & 0xffffffff;
376 cmd->cdw11 = slba >> 32;
377 /* cdw12 represent number of lba's for read/write */
a7e8aae0
KB		 378 cmd->cdw12 = nlb | (io_u->dtype << 20);
379 cmd->cdw13 = io_u->dspec << 16;
b3d5e3fd
AK		 380 if (iov) {
381 iov->iov_base = io_u->xfer_buf;
382 iov->iov_len = io_u->xfer_buflen;
383 cmd->addr = (__u64)(uintptr_t)iov;
384 cmd->data_len = 1;
385 } else {
386 cmd->addr = (__u64)(uintptr_t)io_u->xfer_buf;
387 cmd->data_len = io_u->xfer_buflen;
388 }
2d6451c9
AK		 389 if (data->lba_shift && data->ms) {
390 cmd->metadata = (__u64)(uintptr_t)io_u->mmap_data;
391 cmd->metadata_len = (nlb + 1) * data->ms;
392 }
b3d5e3fd
AK		 393 cmd->nsid = data->nsid;
394 return 0;
395}
396
3ee8311a
AK		 397void fio_nvme_pi_fill(struct nvme_uring_cmd *cmd, struct io_u *io_u,
398 struct nvme_cmd_ext_io_opts *opts)
399{
400 struct nvme_data *data = FILE_ENG_DATA(io_u->file);
401 __u64 slba;
402
403 slba = get_slba(data, io_u);
404 cmd->cdw12 |= opts->io_flags;
405
5163f35e
AK		 406 if (data->pi_type && !(opts->io_flags & NVME_IO_PRINFO_PRACT)) {
407 if (data->guard_type == NVME_NVM_NS_16B_GUARD)
408 fio_nvme_generate_pi_16b_guard(data, io_u, opts);
08371767
AK		 409 else if (data->guard_type == NVME_NVM_NS_64B_GUARD)
410 fio_nvme_generate_pi_64b_guard(data, io_u, opts);
5163f35e
AK		 411 }
412
3ee8311a
AK		 413 switch (data->pi_type) {
414 case NVME_NS_DPS_PI_TYPE1:
415 case NVME_NS_DPS_PI_TYPE2:
416 switch (data->guard_type) {
417 case NVME_NVM_NS_16B_GUARD:
418 cmd->cdw14 = (__u32)slba;
419 break;
420 case NVME_NVM_NS_64B_GUARD:
421 cmd->cdw14 = (__u32)slba;
422 cmd->cdw3 = ((slba >> 32) & 0xffff);
423 break;
424 default:
425 break;
426 }
427 cmd->cdw15 = (opts->apptag_mask << 16 | opts->apptag);
428 break;
429 case NVME_NS_DPS_PI_TYPE3:
430 cmd->cdw15 = (opts->apptag_mask << 16 | opts->apptag);
431 break;
432 case NVME_NS_DPS_PI_NONE:
433 break;
434 }
435}
436
5163f35e
AK		 437int fio_nvme_pi_verify(struct nvme_data *data, struct io_u *io_u)
438{
439 int ret = 0;
440
441 switch (data->guard_type) {
442 case NVME_NVM_NS_16B_GUARD:
443 ret = fio_nvme_verify_pi_16b_guard(data, io_u);
444 break;
08371767
AK		 445 case NVME_NVM_NS_64B_GUARD:
446 ret = fio_nvme_verify_pi_64b_guard(data, io_u);
447 break;
5163f35e
AK		 448 default:
449 break;
450 }
451
452 return ret;
453}
454
b3d5e3fd
AK		 455static int nvme_identify(int fd, __u32 nsid, enum nvme_identify_cns cns,
456 enum nvme_csi csi, void *data)
457{
458 struct nvme_passthru_cmd cmd = {
459 .opcode = nvme_admin_identify,
460 .nsid = nsid,
461 .addr = (__u64)(uintptr_t)data,
462 .data_len = NVME_IDENTIFY_DATA_SIZE,
463 .cdw10 = cns,
464 .cdw11 = csi << NVME_IDENTIFY_CSI_SHIFT,
465 .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT,
466 };
467
468 return ioctl(fd, NVME_IOCTL_ADMIN_CMD, &cmd);
469}
470
3ee8311a
AK		 471int fio_nvme_get_info(struct fio_file *f, __u64 *nlba, __u32 pi_act,
472 struct nvme_data *data)
b3d5e3fd
AK		 473{
474 struct nvme_id_ns ns;
3ee8311a
AK		 475 struct nvme_id_ctrl ctrl;
476 struct nvme_nvm_id_ns nvm_ns;
37a0881f 477 int namespace_id;
b3d5e3fd 478 int fd, err;
3ee8311a 479 __u32 format_idx, elbaf;
b3d5e3fd
AK		 480
481 if (f->filetype != FIO_TYPE_CHAR) {
482 log_err("ioengine io_uring_cmd only works with nvme ns "
483 "generic char devices (/dev/ngXnY)\n");
484 return 1;
485 }
486
487 fd = open(f->file_name, O_RDONLY);
488 if (fd < 0)
489 return -errno;
490
491 namespace_id = ioctl(fd, NVME_IOCTL_ID);
492 if (namespace_id < 0) {
af10f514 493 err = -errno;
345fa8fd
AK		 494 log_err("%s: failed to fetch namespace-id\n", f->file_name);
495 goto out;
b3d5e3fd
AK		 496 }
497
3ee8311a
AK		 498 err = nvme_identify(fd, 0, NVME_IDENTIFY_CNS_CTRL, NVME_CSI_NVM, &ctrl);
499 if (err) {
500 log_err("%s: failed to fetch identify ctrl\n", f->file_name);
501 goto out;
502 }
503
b3d5e3fd
AK		 504 /*
505 * Identify namespace to get namespace-id, namespace size in LBA's
506 * and LBA data size.
507 */
508 err = nvme_identify(fd, namespace_id, NVME_IDENTIFY_CNS_NS,
509 NVME_CSI_NVM, &ns);
510 if (err) {
345fa8fd
AK		 511 log_err("%s: failed to fetch identify namespace\n",
512 f->file_name);
3ee8311a 513 goto out;
b3d5e3fd
AK		 514 }
515
e7e5023b 516 data->nsid = namespace_id;
01a7d384
AK		 517
518 /*
519 * 16 or 64 as maximum number of supported LBA formats.
520 * From flbas bit 0-3 indicates lsb and bit 5-6 indicates msb
521 * of the format index used to format the namespace.
522 */
523 if (ns.nlbaf < 16)
524 format_idx = ns.flbas & 0xf;
525 else
526 format_idx = (ns.flbas & 0xf) + (((ns.flbas >> 5) & 0x3) << 4);
527
e7e5023b 528 data->lba_size = 1 << ns.lbaf[format_idx].ds;
2d6451c9 529 data->ms = le16_to_cpu(ns.lbaf[format_idx].ms);
345fa8fd 530
3ee8311a
AK		 531 /* Check for end to end data protection support */
532 if (data->ms && (ns.dps & NVME_NS_DPS_PI_MASK))
533 data->pi_type = (ns.dps & NVME_NS_DPS_PI_MASK);
534
535 if (!data->pi_type)
536 goto check_elba;
537
538 if (ctrl.ctratt & NVME_CTRL_CTRATT_ELBAS) {
539 err = nvme_identify(fd, namespace_id, NVME_IDENTIFY_CNS_CSI_NS,
540 NVME_CSI_NVM, &nvm_ns);
541 if (err) {
542 log_err("%s: failed to fetch identify nvm namespace\n",
543 f->file_name);
544 goto out;
545 }
546
547 elbaf = le32_to_cpu(nvm_ns.elbaf[format_idx]);
548
549 /* Currently we don't support storage tags */
550 if (elbaf & NVME_ID_NS_NVM_STS_MASK) {
551 log_err("%s: Storage tag not supported\n",
552 f->file_name);
553 err = -ENOTSUP;
554 goto out;
555 }
556
557 data->guard_type = (elbaf >> NVME_ID_NS_NVM_GUARD_SHIFT) &
558 NVME_ID_NS_NVM_GUARD_MASK;
559
560 /* No 32 bit guard, as storage tag is mandatory for it */
561 switch (data->guard_type) {
562 case NVME_NVM_NS_16B_GUARD:
563 data->pi_size = sizeof(struct nvme_16b_guard_pif);
564 break;
565 case NVME_NVM_NS_64B_GUARD:
566 data->pi_size = sizeof(struct nvme_64b_guard_pif);
567 break;
568 default:
569 break;
570 }
571 } else {
572 data->guard_type = NVME_NVM_NS_16B_GUARD;
573 data->pi_size = sizeof(struct nvme_16b_guard_pif);
574 }
575
576 /*
577 * when PRACT bit is set to 1, and metadata size is equal to protection
578 * information size, controller inserts and removes PI for write and
579 * read commands respectively.
580 */
581 if (pi_act && data->ms == data->pi_size)
582 data->ms = 0;
583
584 data->pi_loc = (ns.dps & NVME_NS_DPS_PI_FIRST);
585
586check_elba:
345fa8fd 587 /*
345fa8fd
AK		 588 * Bit 4 for flbas indicates if metadata is transferred at the end of
589 * logical block creating an extended LBA.
590 */
2d6451c9 591 if (data->ms && ((ns.flbas >> 4) & 0x1))
e7e5023b
AK		 592 data->lba_ext = data->lba_size + data->ms;
593 else
594 data->lba_shift = ilog2(data->lba_size);
595
b3d5e3fd
AK		 596 *nlba = ns.nsze;
597
345fa8fd 598out:
b3d5e3fd 599 close(fd);
345fa8fd 600 return err;
b3d5e3fd 601}
3d05e0ff
AK		 602
603int fio_nvme_get_zoned_model(struct thread_data *td, struct fio_file *f,
604 enum zbd_zoned_model *model)
605{
606 struct nvme_data *data = FILE_ENG_DATA(f);
607 struct nvme_id_ns ns;
608 struct nvme_passthru_cmd cmd;
609 int fd, ret = 0;
610
611 if (f->filetype != FIO_TYPE_CHAR)
612 return -EINVAL;
613
614 /* File is not yet opened */
615 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
616 if (fd < 0)
617 return -errno;
618
619 /* Using nvme_id_ns for data as sizes are same */
620 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_CTRL,
621 NVME_CSI_ZNS, &ns);
622 if (ret) {
623 *model = ZBD_NONE;
624 goto out;
625 }
626
627 memset(&cmd, 0, sizeof(struct nvme_passthru_cmd));
628
629 /* Using nvme_id_ns for data as sizes are same */
630 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_NS,
631 NVME_CSI_ZNS, &ns);
632 if (ret) {
633 *model = ZBD_NONE;
634 goto out;
635 }
636
637 *model = ZBD_HOST_MANAGED;
638out:
639 close(fd);
640 return 0;
641}
642
643static int nvme_report_zones(int fd, __u32 nsid, __u64 slba, __u32 zras_feat,
644 __u32 data_len, void *data)
645{
646 struct nvme_passthru_cmd cmd = {
647 .opcode = nvme_zns_cmd_mgmt_recv,
648 .nsid = nsid,
649 .addr = (__u64)(uintptr_t)data,
650 .data_len = data_len,
651 .cdw10 = slba & 0xffffffff,
652 .cdw11 = slba >> 32,
653 .cdw12 = (data_len >> 2) - 1,
654 .cdw13 = NVME_ZNS_ZRA_REPORT_ZONES | zras_feat,
655 .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT,
656 };
657
658 return ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
659}
660
661int fio_nvme_report_zones(struct thread_data *td, struct fio_file *f,
662 uint64_t offset, struct zbd_zone *zbdz,
663 unsigned int nr_zones)
664{
665 struct nvme_data *data = FILE_ENG_DATA(f);
666 struct nvme_zone_report *zr;
667 struct nvme_zns_id_ns zns_ns;
668 struct nvme_id_ns ns;
669 unsigned int i = 0, j, zones_fetched = 0;
670 unsigned int max_zones, zones_chunks = 1024;
671 int fd, ret = 0;
672 __u32 zr_len;
673 __u64 zlen;
674
675 /* File is not yet opened */
676 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
677 if (fd < 0)
678 return -errno;
679
680 zones_fetched = 0;
681 zr_len = sizeof(*zr) + (zones_chunks * sizeof(struct nvme_zns_desc));
682 zr = calloc(1, zr_len);
3efcb23f
JA		 683 if (!zr) {
684 close(fd);
3d05e0ff 685 return -ENOMEM;
3efcb23f 686 }
3d05e0ff
AK		 687
688 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_NS,
689 NVME_CSI_NVM, &ns);
690 if (ret) {
691 log_err("%s: nvme_identify_ns failed, err=%d\n", f->file_name,
692 ret);
693 goto out;
694 }
695
696 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_NS,
697 NVME_CSI_ZNS, &zns_ns);
698 if (ret) {
699 log_err("%s: nvme_zns_identify_ns failed, err=%d\n",
700 f->file_name, ret);
701 goto out;
702 }
703 zlen = zns_ns.lbafe[ns.flbas & 0x0f].zsze << data->lba_shift;
704
705 max_zones = (f->real_file_size - offset) / zlen;
706 if (max_zones < nr_zones)
707 nr_zones = max_zones;
708
709 if (nr_zones < zones_chunks)
710 zones_chunks = nr_zones;
711
712 while (zones_fetched < nr_zones) {
713 if (zones_fetched + zones_chunks >= nr_zones) {
714 zones_chunks = nr_zones - zones_fetched;
715 zr_len = sizeof(*zr) + (zones_chunks * sizeof(struct nvme_zns_desc));
716 }
717 ret = nvme_report_zones(fd, data->nsid, offset >> data->lba_shift,
718 NVME_ZNS_ZRAS_FEAT_ERZ, zr_len, (void *)zr);
719 if (ret) {
720 log_err("%s: nvme_zns_report_zones failed, err=%d\n",
721 f->file_name, ret);
722 goto out;
723 }
724
725 /* Transform the zone-report */
726 for (j = 0; j < zr->nr_zones; j++, i++) {
727 struct nvme_zns_desc *desc = (struct nvme_zns_desc *)&(zr->entries[j]);
728
729 zbdz[i].start = desc->zslba << data->lba_shift;
730 zbdz[i].len = zlen;
731 zbdz[i].wp = desc->wp << data->lba_shift;
732 zbdz[i].capacity = desc->zcap << data->lba_shift;
733
734 /* Zone Type is stored in first 4 bits. */
735 switch (desc->zt & 0x0f) {
736 case NVME_ZONE_TYPE_SEQWRITE_REQ:
737 zbdz[i].type = ZBD_ZONE_TYPE_SWR;
738 break;
739 default:
740 log_err("%s: invalid type for zone at offset %llu.\n",
2fa0ab21 741 f->file_name, (unsigned long long) desc->zslba);
3d05e0ff
AK		 742 ret = -EIO;
743 goto out;
744 }
745
746 /* Zone State is stored in last 4 bits. */
747 switch (desc->zs >> 4) {
748 case NVME_ZNS_ZS_EMPTY:
749 zbdz[i].cond = ZBD_ZONE_COND_EMPTY;
750 break;
751 case NVME_ZNS_ZS_IMPL_OPEN:
752 zbdz[i].cond = ZBD_ZONE_COND_IMP_OPEN;
753 break;
754 case NVME_ZNS_ZS_EXPL_OPEN:
755 zbdz[i].cond = ZBD_ZONE_COND_EXP_OPEN;
756 break;
757 case NVME_ZNS_ZS_CLOSED:
758 zbdz[i].cond = ZBD_ZONE_COND_CLOSED;
759 break;
760 case NVME_ZNS_ZS_FULL:
761 zbdz[i].cond = ZBD_ZONE_COND_FULL;
762 break;
763 case NVME_ZNS_ZS_READ_ONLY:
764 case NVME_ZNS_ZS_OFFLINE:
765 default:
766 /* Treat all these conditions as offline (don't use!) */
767 zbdz[i].cond = ZBD_ZONE_COND_OFFLINE;
768 zbdz[i].wp = zbdz[i].start;
769 }
770 }
771 zones_fetched += zr->nr_zones;
772 offset += zr->nr_zones * zlen;
773 }
774
775 ret = zones_fetched;
776out:
777 free(zr);
778 close(fd);
779
780 return ret;
781}
782
783int fio_nvme_reset_wp(struct thread_data *td, struct fio_file *f,
784 uint64_t offset, uint64_t length)
785{
786 struct nvme_data *data = FILE_ENG_DATA(f);
787 unsigned int nr_zones;
788 unsigned long long zslba;
789 int i, fd, ret = 0;
790
791 /* If the file is not yet opened, open it for this function. */
792 fd = f->fd;
793 if (fd < 0) {
794 fd = open(f->file_name, O_RDWR | O_LARGEFILE);
795 if (fd < 0)
796 return -errno;
797 }
798
799 zslba = offset >> data->lba_shift;
800 nr_zones = (length + td->o.zone_size - 1) / td->o.zone_size;
801
802 for (i = 0; i < nr_zones; i++, zslba += (td->o.zone_size >> data->lba_shift)) {
803 struct nvme_passthru_cmd cmd = {
804 .opcode = nvme_zns_cmd_mgmt_send,
805 .nsid = data->nsid,
806 .cdw10 = zslba & 0xffffffff,
807 .cdw11 = zslba >> 32,
808 .cdw13 = NVME_ZNS_ZSA_RESET,
809 .addr = (__u64)(uintptr_t)NULL,
810 .data_len = 0,
811 .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT,
812 };
813
814 ret = ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
815 }
816
817 if (f->fd < 0)
818 close(fd);
819 return -ret;
820}
821
822int fio_nvme_get_max_open_zones(struct thread_data *td, struct fio_file *f,
823 unsigned int *max_open_zones)
824{
825 struct nvme_data *data = FILE_ENG_DATA(f);
826 struct nvme_zns_id_ns zns_ns;
827 int fd, ret = 0;
828
829 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
830 if (fd < 0)
831 return -errno;
832
833 ret = nvme_identify(fd, data->nsid, NVME_IDENTIFY_CNS_CSI_NS,
834 NVME_CSI_ZNS, &zns_ns);
835 if (ret) {
836 log_err("%s: nvme_zns_identify_ns failed, err=%d\n",
837 f->file_name, ret);
838 goto out;
839 }
840
841 *max_open_zones = zns_ns.mor + 1;
842out:
843 close(fd);
844 return ret;
845}
a7e8aae0
KB		 846
847static inline int nvme_fdp_reclaim_unit_handle_status(int fd, __u32 nsid,
848 __u32 data_len, void *data)
849{
850 struct nvme_passthru_cmd cmd = {
851 .opcode = nvme_cmd_io_mgmt_recv,
852 .nsid = nsid,
853 .addr = (__u64)(uintptr_t)data,
854 .data_len = data_len,
855 .cdw10 = 1,
856 .cdw11 = (data_len >> 2) - 1,
857 };
858
859 return ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
860}
861
862int fio_nvme_iomgmt_ruhs(struct thread_data *td, struct fio_file *f,
863 struct nvme_fdp_ruh_status *ruhs, __u32 bytes)
864{
865 struct nvme_data *data = FILE_ENG_DATA(f);
866 int fd, ret;
867
868 fd = open(f->file_name, O_RDONLY | O_LARGEFILE);
869 if (fd < 0)
870 return -errno;
871
872 ret = nvme_fdp_reclaim_unit_handle_status(fd, data->nsid, bytes, ruhs);
873 if (ret) {
874 log_err("%s: nvme_fdp_reclaim_unit_handle_status failed, err=%d\n",
875 f->file_name, ret);
876 errno = ENOTSUP;
877 } else
878 errno = 0;
879
af10f514 880 ret = -errno;
a7e8aae0 881 close(fd);
af10f514 882 return ret;
a7e8aae0 883}
fio - Flexible IO Tester