Commit | Line | Data |
---|---|---|
bfbdd35b BVA |
1 | /* |
2 | * Copyright (C) 2018 Western Digital Corporation or its affiliates. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include <errno.h> | |
8 | #include <string.h> | |
9 | #include <stdlib.h> | |
bfbdd35b | 10 | #include <fcntl.h> |
bfbdd35b BVA |
11 | #include <sys/stat.h> |
12 | #include <unistd.h> | |
f5bff36e | 13 | |
cf42d79e | 14 | #include "os/os.h" |
bfbdd35b BVA |
15 | #include "file.h" |
16 | #include "fio.h" | |
17 | #include "lib/pow2.h" | |
18 | #include "log.h" | |
f5bff36e | 19 | #include "oslib/asprintf.h" |
bfbdd35b BVA |
20 | #include "smalloc.h" |
21 | #include "verify.h" | |
44ec32cb | 22 | #include "pshared.h" |
bfbdd35b BVA |
23 | #include "zbd.h" |
24 | ||
b7694961 DLM |
25 | /** |
26 | * zbd_get_zoned_model - Get a device zoned model | |
27 | * @td: FIO thread data | |
28 | * @f: FIO file for which to get model information | |
29 | */ | |
30 | int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, | |
31 | enum zbd_zoned_model *model) | |
32 | { | |
33 | int ret; | |
34 | ||
50cc48d5 NC |
35 | if (f->filetype == FIO_TYPE_PIPE) { |
36 | log_err("zonemode=zbd does not support pipes\n"); | |
37 | return -EINVAL; | |
38 | } | |
39 | ||
9db0cde8 NC |
40 | /* If regular file, always emulate zones inside the file. */ |
41 | if (f->filetype == FIO_TYPE_FILE) { | |
42 | *model = ZBD_NONE; | |
43 | return 0; | |
44 | } | |
45 | ||
6c5b11d3 DLM |
46 | if (td->io_ops && td->io_ops->get_zoned_model) |
47 | ret = td->io_ops->get_zoned_model(td, f, model); | |
48 | else | |
49 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
50 | if (ret < 0) { |
51 | td_verror(td, errno, "get zoned model failed"); | |
52 | log_err("%s: get zoned model failed (%d).\n", | |
53 | f->file_name, errno); | |
54 | } | |
55 | ||
56 | return ret; | |
57 | } | |
58 | ||
59 | /** | |
60 | * zbd_report_zones - Get zone information | |
61 | * @td: FIO thread data. | |
62 | * @f: FIO file for which to get zone information | |
63 | * @offset: offset from which to report zones | |
64 | * @zones: Array of struct zbd_zone | |
65 | * @nr_zones: Size of @zones array | |
66 | * | |
67 | * Get zone information into @zones starting from the zone at offset @offset | |
68 | * for the device specified by @f. | |
69 | * | |
70 | * Returns the number of zones reported upon success and a negative error code | |
71 | * upon failure. If the zone report is empty, always assume an error (device | |
72 | * problem) and return -EIO. | |
73 | */ | |
74 | int zbd_report_zones(struct thread_data *td, struct fio_file *f, | |
75 | uint64_t offset, struct zbd_zone *zones, | |
76 | unsigned int nr_zones) | |
77 | { | |
78 | int ret; | |
79 | ||
6c5b11d3 DLM |
80 | if (td->io_ops && td->io_ops->report_zones) |
81 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
82 | else | |
83 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
84 | if (ret < 0) { |
85 | td_verror(td, errno, "report zones failed"); | |
86 | log_err("%s: report zones from sector %llu failed (%d).\n", | |
87 | f->file_name, (unsigned long long)offset >> 9, errno); | |
88 | } else if (ret == 0) { | |
89 | td_verror(td, errno, "Empty zone report"); | |
90 | log_err("%s: report zones from sector %llu is empty.\n", | |
91 | f->file_name, (unsigned long long)offset >> 9); | |
92 | ret = -EIO; | |
93 | } | |
94 | ||
95 | return ret; | |
96 | } | |
97 | ||
98 | /** | |
99 | * zbd_reset_wp - reset the write pointer of a range of zones | |
100 | * @td: FIO thread data. | |
101 | * @f: FIO file for which to reset zones | |
102 | * @offset: Starting offset of the first zone to reset | |
103 | * @length: Length of the range of zones to reset | |
104 | * | |
105 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
106 | * Returns 0 upon success and a negative error code upon failure. | |
107 | */ | |
108 | int zbd_reset_wp(struct thread_data *td, struct fio_file *f, | |
109 | uint64_t offset, uint64_t length) | |
110 | { | |
111 | int ret; | |
112 | ||
6c5b11d3 DLM |
113 | if (td->io_ops && td->io_ops->reset_wp) |
114 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
115 | else | |
116 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
117 | if (ret < 0) { |
118 | td_verror(td, errno, "resetting wp failed"); | |
119 | log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n", | |
120 | f->file_name, (unsigned long long)length >> 9, | |
121 | (unsigned long long)offset >> 9, errno); | |
122 | } | |
123 | ||
124 | return ret; | |
125 | } | |
126 | ||
d2f442bc NC |
127 | /** |
128 | * zbd_get_max_open_zones - Get the maximum number of open zones | |
129 | * @td: FIO thread data | |
130 | * @f: FIO file for which to get max open zones | |
131 | * @max_open_zones: Upon success, result will be stored here. | |
132 | * | |
133 | * A @max_open_zones value set to zero means no limit. | |
134 | * | |
135 | * Returns 0 upon success and a negative error code upon failure. | |
136 | */ | |
137 | int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, | |
138 | unsigned int *max_open_zones) | |
139 | { | |
140 | int ret; | |
141 | ||
142 | if (td->io_ops && td->io_ops->get_max_open_zones) | |
143 | ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); | |
144 | else | |
145 | ret = blkzoned_get_max_open_zones(td, f, max_open_zones); | |
146 | if (ret < 0) { | |
147 | td_verror(td, errno, "get max open zones failed"); | |
148 | log_err("%s: get max open zones failed (%d).\n", | |
149 | f->file_name, errno); | |
150 | } | |
151 | ||
152 | return ret; | |
153 | } | |
154 | ||
bfbdd35b BVA |
155 | /** |
156 | * zbd_zone_idx - convert an offset into a zone number | |
157 | * @f: file pointer. | |
158 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
159 | * past the disk size then the index of the sentinel is returned. | |
160 | */ | |
161 | static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) | |
162 | { | |
163 | uint32_t zone_idx; | |
164 | ||
cd775e06 | 165 | if (f->zbd_info->zone_size_log2 > 0) |
bfbdd35b BVA |
166 | zone_idx = offset >> f->zbd_info->zone_size_log2; |
167 | else | |
ee3696bd | 168 | zone_idx = offset / f->zbd_info->zone_size; |
bfbdd35b BVA |
169 | |
170 | return min(zone_idx, f->zbd_info->nr_zones); | |
171 | } | |
172 | ||
236d23a8 SK |
173 | /** |
174 | * zbd_zone_end - Return zone end location | |
175 | * @z: zone info pointer. | |
176 | */ | |
177 | static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) | |
178 | { | |
179 | return (z+1)->start; | |
180 | } | |
181 | ||
182 | /** | |
183 | * zbd_zone_capacity_end - Return zone capacity limit end location | |
184 | * @z: zone info pointer. | |
185 | */ | |
186 | static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) | |
187 | { | |
188 | return z->start + z->capacity; | |
189 | } | |
190 | ||
bfbdd35b BVA |
191 | /** |
192 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
193 | * @f: file pointer. | |
194 | * @z: zone info pointer. | |
195 | * @required: minimum number of bytes that must remain in a zone. | |
196 | * | |
197 | * The caller must hold z->mutex. | |
198 | */ | |
199 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
200 | uint64_t required) | |
201 | { | |
202 | assert((required & 511) == 0); | |
203 | ||
be7a6bae | 204 | return z->has_wp && |
236d23a8 | 205 | z->wp + required > zbd_zone_capacity_end(z); |
bfbdd35b BVA |
206 | } |
207 | ||
11afb212 DF |
208 | static void zone_lock(struct thread_data *td, const struct fio_file *f, |
209 | struct fio_zone_info *z) | |
1f57803b | 210 | { |
fae3b9a0 AD |
211 | struct zoned_block_device_info *zbd = f->zbd_info; |
212 | uint32_t nz = z - zbd->zone_info; | |
213 | ||
214 | /* A thread should never lock zones outside its working area. */ | |
215 | assert(f->min_zone <= nz && nz < f->max_zone); | |
216 | ||
43bcbd5b SK |
217 | assert(z->has_wp); |
218 | ||
1f57803b DLM |
219 | /* |
220 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
221 | * is changed or when io_u completes and zbd_put_io() executed. | |
222 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
223 | * other waiting for zone locks when building an io_u batch, first | |
224 | * only trylock the zone. If the zone is already locked by another job, | |
225 | * process the currently queued I/Os so that I/O progress is made and | |
226 | * zones unlocked. | |
227 | */ | |
228 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
229 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
230 | io_u_quiesce(td); | |
231 | pthread_mutex_lock(&z->mutex); | |
232 | } | |
233 | } | |
234 | ||
4d4c71e6 DF |
235 | static inline void zone_unlock(struct fio_zone_info *z) |
236 | { | |
237 | int ret; | |
238 | ||
43bcbd5b | 239 | assert(z->has_wp); |
4d4c71e6 DF |
240 | ret = pthread_mutex_unlock(&z->mutex); |
241 | assert(!ret); | |
242 | } | |
243 | ||
bfbdd35b BVA |
244 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
245 | { | |
246 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
247 | } | |
248 | ||
923f7c1e DF |
249 | static inline struct fio_zone_info *get_zone(const struct fio_file *f, |
250 | unsigned int zone_nr) | |
251 | { | |
252 | return &f->zbd_info->zone_info[zone_nr]; | |
253 | } | |
254 | ||
bfbdd35b BVA |
255 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ |
256 | static bool zbd_using_direct_io(void) | |
257 | { | |
258 | struct thread_data *td; | |
259 | struct fio_file *f; | |
260 | int i, j; | |
261 | ||
262 | for_each_td(td, i) { | |
263 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
264 | continue; | |
265 | for_each_file(td, f, j) { | |
266 | if (f->zbd_info && | |
b7694961 | 267 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
268 | return false; |
269 | } | |
270 | } | |
271 | ||
272 | return true; | |
273 | } | |
274 | ||
275 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
276 | static bool zbd_is_seq_job(struct fio_file *f) | |
277 | { | |
278 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
279 | ||
280 | assert(f->zbd_info); | |
281 | if (f->io_size == 0) | |
282 | return false; | |
283 | zone_idx_b = zbd_zone_idx(f, f->file_offset); | |
284 | zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); | |
285 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) | |
be7a6bae | 286 | if (get_zone(f, zone_idx)->has_wp) |
bfbdd35b BVA |
287 | return true; |
288 | ||
289 | return false; | |
290 | } | |
291 | ||
292 | /* | |
293 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
294 | */ | |
295 | static bool zbd_verify_sizes(void) | |
296 | { | |
297 | const struct fio_zone_info *z; | |
298 | struct thread_data *td; | |
299 | struct fio_file *f; | |
300 | uint64_t new_offset, new_end; | |
301 | uint32_t zone_idx; | |
302 | int i, j; | |
303 | ||
304 | for_each_td(td, i) { | |
305 | for_each_file(td, f, j) { | |
306 | if (!f->zbd_info) | |
307 | continue; | |
308 | if (f->file_offset >= f->real_file_size) | |
309 | continue; | |
310 | if (!zbd_is_seq_job(f)) | |
311 | continue; | |
4d37720a | 312 | |
fc52e449 DLM |
313 | if (!td->o.zone_size) { |
314 | td->o.zone_size = f->zbd_info->zone_size; | |
315 | if (!td->o.zone_size) { | |
316 | log_err("%s: invalid 0 zone size\n", | |
317 | f->file_name); | |
318 | return false; | |
319 | } | |
971d6a22 DLM |
320 | } else if (td->o.zone_size != f->zbd_info->zone_size) { |
321 | log_err("%s: job parameter zonesize %llu does not match disk zone size %llu.\n", | |
322 | f->file_name, (unsigned long long) td->o.zone_size, | |
323 | (unsigned long long) f->zbd_info->zone_size); | |
324 | return false; | |
fc52e449 DLM |
325 | } |
326 | ||
4316f640 | 327 | if (td->o.zone_skip % td->o.zone_size) { |
4d37720a DLM |
328 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", |
329 | f->file_name, (unsigned long long) td->o.zone_skip, | |
330 | (unsigned long long) td->o.zone_size); | |
331 | return false; | |
332 | } | |
333 | ||
bfbdd35b | 334 | zone_idx = zbd_zone_idx(f, f->file_offset); |
923f7c1e | 335 | z = get_zone(f, zone_idx); |
f09a7773 PL |
336 | if ((f->file_offset != z->start) && |
337 | (td->o.td_ddir != TD_DDIR_READ)) { | |
236d23a8 | 338 | new_offset = zbd_zone_end(z); |
bfbdd35b BVA |
339 | if (new_offset >= f->file_offset + f->io_size) { |
340 | log_info("%s: io_size must be at least one zone\n", | |
341 | f->file_name); | |
342 | return false; | |
343 | } | |
a0c84dd4 JA |
344 | log_info("%s: rounded up offset from %llu to %llu\n", |
345 | f->file_name, (unsigned long long) f->file_offset, | |
346 | (unsigned long long) new_offset); | |
bfbdd35b BVA |
347 | f->io_size -= (new_offset - f->file_offset); |
348 | f->file_offset = new_offset; | |
349 | } | |
350 | zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); | |
923f7c1e | 351 | z = get_zone(f, zone_idx); |
ee3696bd | 352 | new_end = z->start; |
f09a7773 PL |
353 | if ((td->o.td_ddir != TD_DDIR_READ) && |
354 | (f->file_offset + f->io_size != new_end)) { | |
bfbdd35b BVA |
355 | if (new_end <= f->file_offset) { |
356 | log_info("%s: io_size must be at least one zone\n", | |
357 | f->file_name); | |
358 | return false; | |
359 | } | |
a0c84dd4 JA |
360 | log_info("%s: rounded down io_size from %llu to %llu\n", |
361 | f->file_name, (unsigned long long) f->io_size, | |
362 | (unsigned long long) new_end - f->file_offset); | |
bfbdd35b BVA |
363 | f->io_size = new_end - f->file_offset; |
364 | } | |
365 | } | |
366 | } | |
367 | ||
368 | return true; | |
369 | } | |
370 | ||
371 | static bool zbd_verify_bs(void) | |
372 | { | |
373 | struct thread_data *td; | |
374 | struct fio_file *f; | |
bfbdd35b BVA |
375 | int i, j, k; |
376 | ||
377 | for_each_td(td, i) { | |
378 | for_each_file(td, f, j) { | |
1ddd225e AD |
379 | uint64_t zone_size; |
380 | ||
bfbdd35b BVA |
381 | if (!f->zbd_info) |
382 | continue; | |
383 | zone_size = f->zbd_info->zone_size; | |
59f94d26 | 384 | for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) { |
bfbdd35b | 385 | if (td->o.verify != VERIFY_NONE && |
ee3696bd | 386 | zone_size % td->o.bs[k] != 0) { |
1ddd225e | 387 | log_info("%s: block size %llu is not a divisor of the zone size %llu\n", |
bfbdd35b | 388 | f->file_name, td->o.bs[k], |
1ddd225e | 389 | (unsigned long long)zone_size); |
bfbdd35b BVA |
390 | return false; |
391 | } | |
392 | } | |
393 | } | |
394 | } | |
395 | return true; | |
396 | } | |
397 | ||
bfbdd35b BVA |
398 | static int ilog2(uint64_t i) |
399 | { | |
400 | int log = -1; | |
401 | ||
402 | while (i) { | |
403 | i >>= 1; | |
404 | log++; | |
405 | } | |
406 | return log; | |
407 | } | |
408 | ||
409 | /* | |
410 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
411 | * allows to execute a ZBD workload against a non-ZBD device. | |
412 | */ | |
413 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
414 | { | |
415 | uint32_t nr_zones; | |
416 | struct fio_zone_info *p; | |
a4b7f12b | 417 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 418 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 419 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
420 | int i; |
421 | ||
a4b7f12b | 422 | if (zone_size == 0) { |
9db0cde8 | 423 | log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", |
a4b7f12b DLM |
424 | f->file_name); |
425 | return 1; | |
426 | } | |
427 | ||
428 | if (zone_size < 512) { | |
429 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
430 | f->file_name); | |
431 | return 1; | |
432 | } | |
433 | ||
b8dd9750 HH |
434 | if (zone_capacity == 0) |
435 | zone_capacity = zone_size; | |
436 | ||
437 | if (zone_capacity > zone_size) { | |
438 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
439 | f->file_name, (unsigned long long) td->o.zone_capacity, | |
440 | (unsigned long long) td->o.zone_size); | |
441 | return 1; | |
442 | } | |
443 | ||
9db0cde8 NC |
444 | if (f->real_file_size < zone_size) { |
445 | log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", | |
446 | f->file_name, f->real_file_size, zone_size); | |
447 | return -EINVAL; | |
448 | } | |
449 | ||
ee3696bd | 450 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
451 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
452 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
453 | if (!zbd_info) | |
454 | return -ENOMEM; | |
455 | ||
44ec32cb | 456 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
457 | zbd_info->refcount = 1; |
458 | p = &zbd_info->zone_info[0]; | |
459 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
460 | mutex_init_pshared_with_type(&p->mutex, |
461 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 462 | p->start = i * zone_size; |
b14651a2 | 463 | p->wp = p->start; |
b7694961 DLM |
464 | p->type = ZBD_ZONE_TYPE_SWR; |
465 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 466 | p->capacity = zone_capacity; |
be7a6bae | 467 | p->has_wp = 1; |
bfbdd35b BVA |
468 | } |
469 | /* a sentinel */ | |
470 | p->start = nr_zones * zone_size; | |
471 | ||
472 | f->zbd_info = zbd_info; | |
473 | f->zbd_info->zone_size = zone_size; | |
474 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 475 | ilog2(zone_size) : 0; |
bfbdd35b | 476 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
477 | return 0; |
478 | } | |
479 | ||
480 | /* | |
b7694961 DLM |
481 | * Maximum number of zones to report in one operation. |
482 | */ | |
483 | #define ZBD_REPORT_MAX_ZONES 8192U | |
484 | ||
485 | /* | |
486 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
487 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
488 | */ |
489 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
490 | { | |
b7694961 DLM |
491 | int nr_zones, nrz; |
492 | struct zbd_zone *zones, *z; | |
bfbdd35b | 493 | struct fio_zone_info *p; |
b7694961 | 494 | uint64_t zone_size, offset; |
bfbdd35b | 495 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 496 | int i, j, ret = -ENOMEM; |
bfbdd35b | 497 | |
b7694961 DLM |
498 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
499 | if (!zones) | |
bfbdd35b BVA |
500 | goto out; |
501 | ||
b7694961 DLM |
502 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
503 | if (nrz < 0) { | |
504 | ret = nrz; | |
505 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
506 | f->file_name, -ret); | |
507 | goto out; | |
bfbdd35b BVA |
508 | } |
509 | ||
b7694961 | 510 | zone_size = zones[0].len; |
ee3696bd | 511 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
512 | |
513 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
514 | td->o.zone_size = zone_size; |
515 | } else if (td->o.zone_size != zone_size) { | |
4580ff92 DLM |
516 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n", |
517 | f->file_name, (unsigned long long) td->o.zone_size, | |
a0c84dd4 | 518 | (unsigned long long) zone_size); |
bfbdd35b | 519 | ret = -EINVAL; |
b7694961 | 520 | goto out; |
bfbdd35b BVA |
521 | } |
522 | ||
a0c84dd4 JA |
523 | dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name, |
524 | nr_zones, (unsigned long long) zone_size / 1024); | |
bfbdd35b BVA |
525 | |
526 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
527 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 528 | if (!zbd_info) |
b7694961 | 529 | goto out; |
44ec32cb | 530 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
531 | zbd_info->refcount = 1; |
532 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
533 | for (offset = 0, j = 0; j < nr_zones;) { |
534 | z = &zones[0]; | |
535 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
536 | mutex_init_pshared_with_type(&p->mutex, |
537 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 538 | p->start = z->start; |
236d23a8 | 539 | p->capacity = z->capacity; |
bfbdd35b | 540 | switch (z->cond) { |
b7694961 DLM |
541 | case ZBD_ZONE_COND_NOT_WP: |
542 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 543 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
544 | break; |
545 | default: | |
546 | assert(z->start <= z->wp); | |
b7694961 DLM |
547 | assert(z->wp <= z->start + zone_size); |
548 | p->wp = z->wp; | |
bfbdd35b BVA |
549 | break; |
550 | } | |
be7a6bae DF |
551 | |
552 | switch (z->type) { | |
553 | case ZBD_ZONE_TYPE_SWR: | |
554 | p->has_wp = 1; | |
555 | break; | |
556 | default: | |
557 | p->has_wp = 0; | |
558 | } | |
bfbdd35b BVA |
559 | p->type = z->type; |
560 | p->cond = z->cond; | |
be7a6bae | 561 | |
bfbdd35b BVA |
562 | if (j > 0 && p->start != p[-1].start + zone_size) { |
563 | log_info("%s: invalid zone data\n", | |
564 | f->file_name); | |
565 | ret = -EINVAL; | |
b7694961 | 566 | goto out; |
bfbdd35b BVA |
567 | } |
568 | } | |
569 | z--; | |
b7694961 | 570 | offset = z->start + z->len; |
bfbdd35b BVA |
571 | if (j >= nr_zones) |
572 | break; | |
6c3f1cc1 DF |
573 | nrz = zbd_report_zones(td, f, offset, zones, |
574 | min((uint32_t)(nr_zones - j), | |
575 | ZBD_REPORT_MAX_ZONES)); | |
b7694961 DLM |
576 | if (nrz < 0) { |
577 | ret = nrz; | |
578 | log_info("fio: report zones (offset %llu) failed for %s (%d).\n", | |
579 | (unsigned long long)offset, | |
580 | f->file_name, -ret); | |
581 | goto out; | |
bfbdd35b BVA |
582 | } |
583 | } | |
b7694961 | 584 | |
bfbdd35b | 585 | /* a sentinel */ |
b7694961 | 586 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
587 | |
588 | f->zbd_info = zbd_info; | |
589 | f->zbd_info->zone_size = zone_size; | |
590 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 591 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
592 | f->zbd_info->nr_zones = nr_zones; |
593 | zbd_info = NULL; | |
594 | ret = 0; | |
595 | ||
bfbdd35b | 596 | out: |
b7694961 DLM |
597 | sfree(zbd_info); |
598 | free(zones); | |
bfbdd35b BVA |
599 | return ret; |
600 | } | |
601 | ||
d2f442bc NC |
602 | static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f) |
603 | { | |
604 | struct zoned_block_device_info *zbd = f->zbd_info; | |
605 | unsigned int max_open_zones; | |
606 | int ret; | |
607 | ||
575686bb | 608 | if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { |
d2f442bc NC |
609 | /* Only host-managed devices have a max open limit */ |
610 | zbd->max_open_zones = td->o.max_open_zones; | |
611 | goto out; | |
612 | } | |
613 | ||
614 | /* If host-managed, get the max open limit */ | |
615 | ret = zbd_get_max_open_zones(td, f, &max_open_zones); | |
616 | if (ret) | |
617 | return ret; | |
618 | ||
619 | if (!max_open_zones) { | |
620 | /* No device limit */ | |
621 | zbd->max_open_zones = td->o.max_open_zones; | |
622 | } else if (!td->o.max_open_zones) { | |
623 | /* No user limit. Set limit to device limit */ | |
624 | zbd->max_open_zones = max_open_zones; | |
625 | } else if (td->o.max_open_zones <= max_open_zones) { | |
626 | /* Both user limit and dev limit. User limit not too large */ | |
627 | zbd->max_open_zones = td->o.max_open_zones; | |
628 | } else { | |
629 | /* Both user limit and dev limit. User limit too large */ | |
630 | td_verror(td, EINVAL, | |
631 | "Specified --max_open_zones is too large"); | |
632 | log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", | |
633 | td->o.max_open_zones, max_open_zones); | |
634 | return -EINVAL; | |
635 | } | |
636 | ||
637 | out: | |
638 | /* Ensure that the limit is not larger than FIO's internal limit */ | |
639 | zbd->max_open_zones = min_not_zero(zbd->max_open_zones, | |
640 | (uint32_t) ZBD_MAX_OPEN_ZONES); | |
641 | dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n", | |
642 | f->file_name, zbd->max_open_zones); | |
643 | ||
644 | return 0; | |
645 | } | |
646 | ||
bfbdd35b BVA |
647 | /* |
648 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
649 | * | |
650 | * Returns 0 upon success and a negative error code upon failure. | |
651 | */ | |
379e5f09 | 652 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 653 | { |
b7694961 DLM |
654 | enum zbd_zoned_model zbd_model; |
655 | int ret; | |
bfbdd35b BVA |
656 | |
657 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
658 | ||
b7694961 DLM |
659 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
660 | if (ret) | |
661 | return ret; | |
662 | ||
bfbdd35b | 663 | switch (zbd_model) { |
b7694961 DLM |
664 | case ZBD_HOST_AWARE: |
665 | case ZBD_HOST_MANAGED: | |
bfbdd35b | 666 | ret = parse_zone_info(td, f); |
d2f442bc NC |
667 | if (ret) |
668 | return ret; | |
bfbdd35b | 669 | break; |
b7694961 | 670 | case ZBD_NONE: |
bfbdd35b | 671 | ret = init_zone_info(td, f); |
d2f442bc NC |
672 | if (ret) |
673 | return ret; | |
bfbdd35b | 674 | break; |
b7694961 DLM |
675 | default: |
676 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
677 | log_err("Unsupported zoned model\n"); | |
678 | return -EINVAL; | |
bfbdd35b | 679 | } |
b7694961 | 680 | |
2c7dd23e | 681 | assert(f->zbd_info); |
d2f442bc NC |
682 | f->zbd_info->model = zbd_model; |
683 | ||
684 | ret = zbd_set_max_open_zones(td, f); | |
685 | if (ret) { | |
686 | zbd_free_zone_info(f); | |
687 | return ret; | |
219c662d | 688 | } |
d2f442bc NC |
689 | |
690 | return 0; | |
bfbdd35b BVA |
691 | } |
692 | ||
693 | void zbd_free_zone_info(struct fio_file *f) | |
694 | { | |
695 | uint32_t refcount; | |
696 | ||
3c1dc34c | 697 | assert(f->zbd_info); |
bfbdd35b BVA |
698 | |
699 | pthread_mutex_lock(&f->zbd_info->mutex); | |
700 | refcount = --f->zbd_info->refcount; | |
701 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
702 | ||
703 | assert((int32_t)refcount >= 0); | |
704 | if (refcount == 0) | |
705 | sfree(f->zbd_info); | |
706 | f->zbd_info = NULL; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Initialize f->zbd_info. | |
711 | * | |
712 | * Returns 0 upon success and a negative error code upon failure. | |
713 | * | |
714 | * Note: this function can only work correctly if it is called before the first | |
715 | * fio fork() call. | |
716 | */ | |
717 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
718 | { | |
719 | struct thread_data *td2; | |
720 | struct fio_file *f2; | |
721 | int i, j, ret; | |
722 | ||
723 | for_each_td(td2, i) { | |
724 | for_each_file(td2, f2, j) { | |
725 | if (td2 == td && f2 == file) | |
726 | continue; | |
727 | if (!f2->zbd_info || | |
728 | strcmp(f2->file_name, file->file_name) != 0) | |
729 | continue; | |
730 | file->zbd_info = f2->zbd_info; | |
731 | file->zbd_info->refcount++; | |
732 | return 0; | |
733 | } | |
734 | } | |
735 | ||
736 | ret = zbd_create_zone_info(td, file); | |
737 | if (ret < 0) | |
c5837eec | 738 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
bfbdd35b BVA |
739 | return ret; |
740 | } | |
741 | ||
954217b9 SK |
742 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
743 | uint32_t zone_idx); | |
744 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
745 | struct fio_zone_info *z); | |
746 | ||
8f39afa7 | 747 | int zbd_init_files(struct thread_data *td) |
bfbdd35b BVA |
748 | { |
749 | struct fio_file *f; | |
750 | int i; | |
751 | ||
752 | for_each_file(td, f, i) { | |
a4b7f12b | 753 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 754 | return 1; |
bfbdd35b | 755 | } |
8f39afa7 AD |
756 | return 0; |
757 | } | |
758 | ||
759 | void zbd_recalc_options_with_zone_granularity(struct thread_data *td) | |
760 | { | |
761 | struct fio_file *f; | |
762 | int i; | |
763 | ||
764 | for_each_file(td, f, i) { | |
765 | struct zoned_block_device_info *zbd = f->zbd_info; | |
766 | // zonemode=strided doesn't get per-file zone size. | |
767 | uint64_t zone_size = zbd ? zbd->zone_size : td->o.zone_size; | |
768 | ||
769 | if (zone_size == 0) | |
770 | continue; | |
771 | ||
772 | if (td->o.size_nz > 0) { | |
773 | td->o.size = td->o.size_nz * zone_size; | |
774 | } | |
775 | if (td->o.io_size_nz > 0) { | |
776 | td->o.io_size = td->o.io_size_nz * zone_size; | |
777 | } | |
778 | if (td->o.start_offset_nz > 0) { | |
779 | td->o.start_offset = td->o.start_offset_nz * zone_size; | |
780 | } | |
781 | if (td->o.offset_increment_nz > 0) { | |
782 | td->o.offset_increment = td->o.offset_increment_nz * zone_size; | |
783 | } | |
784 | if (td->o.zone_skip_nz > 0) { | |
785 | td->o.zone_skip = td->o.zone_skip_nz * zone_size; | |
786 | } | |
787 | } | |
788 | } | |
789 | ||
790 | int zbd_setup_files(struct thread_data *td) | |
791 | { | |
792 | struct fio_file *f; | |
793 | int i; | |
bfbdd35b BVA |
794 | |
795 | if (!zbd_using_direct_io()) { | |
796 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
797 | return 1; | |
798 | } | |
799 | ||
800 | if (!zbd_verify_sizes()) | |
801 | return 1; | |
802 | ||
803 | if (!zbd_verify_bs()) | |
804 | return 1; | |
805 | ||
219c662d AD |
806 | for_each_file(td, f, i) { |
807 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
808 | struct fio_zone_info *z; |
809 | int zi; | |
219c662d | 810 | |
5ddf46d0 | 811 | assert(zbd); |
219c662d | 812 | |
f952800a SK |
813 | f->min_zone = zbd_zone_idx(f, f->file_offset); |
814 | f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size); | |
815 | ||
816 | /* | |
817 | * When all zones in the I/O range are conventional, io_size | |
818 | * can be smaller than zone size, making min_zone the same | |
819 | * as max_zone. This is why the assert below needs to be made | |
820 | * conditional. | |
821 | */ | |
822 | if (zbd_is_seq_job(f)) | |
823 | assert(f->min_zone < f->max_zone); | |
824 | ||
219c662d AD |
825 | if (td->o.max_open_zones > 0 && |
826 | zbd->max_open_zones != td->o.max_open_zones) { | |
827 | log_err("Different 'max_open_zones' values\n"); | |
828 | return 1; | |
829 | } | |
830 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { | |
831 | log_err("'max_open_zones' value is limited by %u\n", ZBD_MAX_OPEN_ZONES); | |
832 | return 1; | |
833 | } | |
954217b9 SK |
834 | |
835 | for (zi = f->min_zone; zi < f->max_zone; zi++) { | |
836 | z = &zbd->zone_info[zi]; | |
837 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
838 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
839 | continue; | |
840 | if (zbd_open_zone(td, f, zi)) | |
841 | continue; | |
842 | /* | |
843 | * If the number of open zones exceeds specified limits, | |
844 | * reset all extra open zones. | |
845 | */ | |
846 | if (zbd_reset_zone(td, f, z) < 0) { | |
847 | log_err("Failed to reest zone %d\n", zi); | |
848 | return 1; | |
849 | } | |
850 | } | |
219c662d AD |
851 | } |
852 | ||
bfbdd35b BVA |
853 | return 0; |
854 | } | |
855 | ||
122c3539 DF |
856 | static inline unsigned int zbd_zone_nr(const struct fio_file *f, |
857 | struct fio_zone_info *zone) | |
ceb35a11 | 858 | { |
122c3539 | 859 | return zone - f->zbd_info->zone_info; |
ceb35a11 DF |
860 | } |
861 | ||
bfbdd35b | 862 | /** |
ceb35a11 | 863 | * zbd_reset_zone - reset the write pointer of a single zone |
bfbdd35b | 864 | * @td: FIO thread data. |
ceb35a11 DF |
865 | * @f: FIO file associated with the disk for which to reset a write pointer. |
866 | * @z: Zone to reset. | |
bfbdd35b BVA |
867 | * |
868 | * Returns 0 upon success and a negative error code upon failure. | |
ceb35a11 DF |
869 | * |
870 | * The caller must hold z->mutex. | |
bfbdd35b | 871 | */ |
ceb35a11 DF |
872 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, |
873 | struct fio_zone_info *z) | |
bfbdd35b | 874 | { |
ceb35a11 DF |
875 | uint64_t offset = z->start; |
876 | uint64_t length = (z+1)->start - offset; | |
403ea795 | 877 | uint64_t data_in_zone = z->wp - z->start; |
bfbdd35b BVA |
878 | int ret = 0; |
879 | ||
403ea795 | 880 | if (!data_in_zone) |
071c9d29 SK |
881 | return 0; |
882 | ||
ee3696bd | 883 | assert(is_valid_offset(f, offset + length - 1)); |
b7694961 | 884 | |
ceb35a11 | 885 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name, |
122c3539 | 886 | zbd_zone_nr(f, z)); |
bfbdd35b | 887 | switch (f->zbd_info->model) { |
b7694961 DLM |
888 | case ZBD_HOST_AWARE: |
889 | case ZBD_HOST_MANAGED: | |
890 | ret = zbd_reset_wp(td, f, offset, length); | |
891 | if (ret < 0) | |
bfbdd35b | 892 | return ret; |
bfbdd35b | 893 | break; |
b7694961 | 894 | default: |
bfbdd35b BVA |
895 | break; |
896 | } | |
897 | ||
ceb35a11 | 898 | pthread_mutex_lock(&f->zbd_info->mutex); |
403ea795 SK |
899 | f->zbd_info->sectors_with_data -= data_in_zone; |
900 | f->zbd_info->wp_sectors_with_data -= data_in_zone; | |
ceb35a11 DF |
901 | pthread_mutex_unlock(&f->zbd_info->mutex); |
902 | z->wp = z->start; | |
903 | z->verify_block = 0; | |
bfbdd35b | 904 | |
ceb35a11 | 905 | td->ts.nr_zone_resets++; |
fd5d733f | 906 | |
bfbdd35b BVA |
907 | return ret; |
908 | } | |
909 | ||
25dc6606 AD |
910 | /* The caller must hold f->zbd_info->mutex */ |
911 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
b2da58c4 | 912 | unsigned int zone_idx) |
25dc6606 | 913 | { |
b2da58c4 | 914 | uint32_t open_zone_idx = 0; |
25dc6606 | 915 | |
b2da58c4 SK |
916 | for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) { |
917 | if (f->zbd_info->open_zones[open_zone_idx] == zone_idx) | |
918 | break; | |
919 | } | |
9fc5c3fd | 920 | if (open_zone_idx == f->zbd_info->num_open_zones) |
b2da58c4 | 921 | return; |
b2da58c4 SK |
922 | |
923 | dprint(FD_ZBD, "%s: closing zone %d\n", f->file_name, zone_idx); | |
25dc6606 AD |
924 | memmove(f->zbd_info->open_zones + open_zone_idx, |
925 | f->zbd_info->open_zones + open_zone_idx + 1, | |
926 | (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) * | |
927 | sizeof(f->zbd_info->open_zones[0])); | |
928 | f->zbd_info->num_open_zones--; | |
219c662d | 929 | td->num_open_zones--; |
923f7c1e | 930 | get_zone(f, zone_idx)->open = 0; |
25dc6606 AD |
931 | } |
932 | ||
bfbdd35b BVA |
933 | /* |
934 | * Reset a range of zones. Returns 0 upon success and 1 upon failure. | |
935 | * @td: fio thread data. | |
936 | * @f: fio file for which to reset zones | |
937 | * @zb: first zone to reset. | |
938 | * @ze: first zone not to reset. | |
bfbdd35b BVA |
939 | */ |
940 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
941 | struct fio_zone_info *const zb, | |
57f88219 | 942 | struct fio_zone_info *const ze) |
bfbdd35b | 943 | { |
4803b841 | 944 | struct fio_zone_info *z; |
ee3696bd | 945 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b BVA |
946 | int res = 0; |
947 | ||
adc6adcb DLM |
948 | assert(min_bs); |
949 | ||
a0c84dd4 | 950 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name, |
122c3539 | 951 | zbd_zone_nr(f, zb), zbd_zone_nr(f, ze)); |
bfbdd35b | 952 | for (z = zb; z < ze; z++) { |
122c3539 | 953 | uint32_t nz = zbd_zone_nr(f, z); |
25dc6606 | 954 | |
be7a6bae | 955 | if (!z->has_wp) |
1f57803b | 956 | continue; |
fae3b9a0 | 957 | zone_lock(td, f, z); |
57f88219 SK |
958 | pthread_mutex_lock(&f->zbd_info->mutex); |
959 | zbd_close_zone(td, f, nz); | |
960 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
961 | if (z->wp != z->start) { | |
1f57803b | 962 | dprint(FD_ZBD, "%s: resetting zone %u\n", |
122c3539 | 963 | f->file_name, zbd_zone_nr(f, z)); |
1f57803b DLM |
964 | if (zbd_reset_zone(td, f, z) < 0) |
965 | res = 1; | |
bfbdd35b | 966 | } |
4d4c71e6 | 967 | zone_unlock(z); |
4803b841 | 968 | } |
bfbdd35b BVA |
969 | |
970 | return res; | |
971 | } | |
972 | ||
a7c2b6fc BVA |
973 | /* |
974 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
975 | * zone reset. | |
976 | */ | |
1bb1bcad AD |
977 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
978 | const struct fio_file *f) | |
a7c2b6fc BVA |
979 | { |
980 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
981 | ||
a7c2b6fc BVA |
982 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
983 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
984 | } |
985 | ||
986 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
987 | const struct fio_file *f) | |
988 | { | |
989 | pthread_mutex_lock(&f->zbd_info->mutex); | |
990 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
991 | pthread_mutex_unlock(&f->zbd_info->mutex); |
992 | } | |
993 | ||
994 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
995 | const struct fio_file *f) | |
996 | { | |
997 | uint32_t write_cnt = 0; | |
998 | ||
999 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1000 | assert(f->zbd_info->write_cnt); | |
1001 | if (f->zbd_info->write_cnt) | |
1002 | write_cnt = --f->zbd_info->write_cnt; | |
1003 | if (write_cnt == 0) | |
1bb1bcad | 1004 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
1005 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1006 | ||
1007 | return write_cnt == 0; | |
1008 | } | |
1009 | ||
91d25131 BVA |
1010 | enum swd_action { |
1011 | CHECK_SWD, | |
1012 | SET_SWD, | |
1013 | }; | |
1014 | ||
1015 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
11afb212 DF |
1016 | static uint64_t zbd_process_swd(struct thread_data *td, |
1017 | const struct fio_file *f, enum swd_action a) | |
615555bb | 1018 | { |
615555bb BVA |
1019 | struct fio_zone_info *zb, *ze, *z; |
1020 | uint64_t swd = 0; | |
403ea795 | 1021 | uint64_t wp_swd = 0; |
615555bb | 1022 | |
923f7c1e DF |
1023 | zb = get_zone(f, f->min_zone); |
1024 | ze = get_zone(f, f->max_zone); | |
615555bb | 1025 | for (z = zb; z < ze; z++) { |
403ea795 | 1026 | if (z->has_wp) { |
11afb212 | 1027 | zone_lock(td, f, z); |
403ea795 SK |
1028 | wp_swd += z->wp - z->start; |
1029 | } | |
615555bb BVA |
1030 | swd += z->wp - z->start; |
1031 | } | |
1032 | pthread_mutex_lock(&f->zbd_info->mutex); | |
91d25131 BVA |
1033 | switch (a) { |
1034 | case CHECK_SWD: | |
1035 | assert(f->zbd_info->sectors_with_data == swd); | |
403ea795 | 1036 | assert(f->zbd_info->wp_sectors_with_data == wp_swd); |
91d25131 BVA |
1037 | break; |
1038 | case SET_SWD: | |
1039 | f->zbd_info->sectors_with_data = swd; | |
403ea795 | 1040 | f->zbd_info->wp_sectors_with_data = wp_swd; |
91d25131 BVA |
1041 | break; |
1042 | } | |
615555bb BVA |
1043 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1044 | for (z = zb; z < ze; z++) | |
403ea795 SK |
1045 | if (z->has_wp) |
1046 | zone_unlock(z); | |
91d25131 BVA |
1047 | |
1048 | return swd; | |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * The swd check is useful for debugging but takes too much time to leave | |
1053 | * it enabled all the time. Hence it is disabled by default. | |
1054 | */ | |
1055 | static const bool enable_check_swd = false; | |
1056 | ||
403ea795 | 1057 | /* Check whether the values of zbd_info.*sectors_with_data are correct. */ |
11afb212 | 1058 | static void zbd_check_swd(struct thread_data *td, const struct fio_file *f) |
91d25131 BVA |
1059 | { |
1060 | if (!enable_check_swd) | |
1061 | return; | |
1062 | ||
11afb212 | 1063 | zbd_process_swd(td, f, CHECK_SWD); |
615555bb BVA |
1064 | } |
1065 | ||
bfbdd35b BVA |
1066 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
1067 | { | |
91d25131 | 1068 | struct fio_zone_info *zb, *ze; |
11afb212 | 1069 | uint64_t swd; |
bfbdd35b | 1070 | |
767d1372 | 1071 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
1072 | return; |
1073 | ||
923f7c1e DF |
1074 | zb = get_zone(f, f->min_zone); |
1075 | ze = get_zone(f, f->max_zone); | |
11afb212 DF |
1076 | swd = zbd_process_swd(td, f, SET_SWD); |
1077 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name, | |
1078 | swd); | |
bfbdd35b BVA |
1079 | /* |
1080 | * If data verification is enabled reset the affected zones before | |
1081 | * writing any data to avoid that a zone reset has to be issued while | |
1082 | * writing data, which causes data loss. | |
1083 | */ | |
57f88219 SK |
1084 | if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING) |
1085 | zbd_reset_zones(td, f, zb, ze); | |
a7c2b6fc | 1086 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
1087 | } |
1088 | ||
59b07544 BVA |
1089 | /* The caller must hold f->zbd_info->mutex. */ |
1090 | static bool is_zone_open(const struct thread_data *td, const struct fio_file *f, | |
1091 | unsigned int zone_idx) | |
1092 | { | |
1093 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
1094 | int i; | |
1095 | ||
219c662d AD |
1096 | assert(td->o.job_max_open_zones == 0 || td->num_open_zones <= td->o.job_max_open_zones); |
1097 | assert(td->o.job_max_open_zones <= zbdi->max_open_zones); | |
1098 | assert(zbdi->num_open_zones <= zbdi->max_open_zones); | |
59b07544 BVA |
1099 | |
1100 | for (i = 0; i < zbdi->num_open_zones; i++) | |
1101 | if (zbdi->open_zones[i] == zone_idx) | |
1102 | return true; | |
1103 | ||
1104 | return false; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Open a ZBD zone if it was not yet open. Returns true if either the zone was | |
1109 | * already open or if opening a new zone is allowed. Returns false if the zone | |
1110 | * was not yet open and opening a new zone would cause the zone limit to be | |
1111 | * exceeded. | |
1112 | */ | |
954217b9 | 1113 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
59b07544 BVA |
1114 | uint32_t zone_idx) |
1115 | { | |
1116 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
923f7c1e | 1117 | struct fio_zone_info *z = get_zone(f, zone_idx); |
59b07544 BVA |
1118 | bool res = true; |
1119 | ||
b7694961 | 1120 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
59b07544 BVA |
1121 | return false; |
1122 | ||
1123 | /* | |
1124 | * Skip full zones with data verification enabled because resetting a | |
1125 | * zone causes data loss and hence causes verification to fail. | |
1126 | */ | |
1127 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) | |
1128 | return false; | |
1129 | ||
59b07544 | 1130 | pthread_mutex_lock(&f->zbd_info->mutex); |
b2da58c4 SK |
1131 | if (is_zone_open(td, f, zone_idx)) { |
1132 | /* | |
1133 | * If the zone is already open and going to be full by writes | |
1134 | * in-flight, handle it as a full zone instead of an open zone. | |
1135 | */ | |
1136 | if (z->wp >= zbd_zone_capacity_end(z)) | |
1137 | res = false; | |
59b07544 | 1138 | goto out; |
b2da58c4 | 1139 | } |
59b07544 | 1140 | res = false; |
219c662d AD |
1141 | /* Zero means no limit */ |
1142 | if (td->o.job_max_open_zones > 0 && | |
1143 | td->num_open_zones >= td->o.job_max_open_zones) | |
1144 | goto out; | |
1145 | if (f->zbd_info->num_open_zones >= f->zbd_info->max_open_zones) | |
59b07544 BVA |
1146 | goto out; |
1147 | dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx); | |
1148 | f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx; | |
219c662d | 1149 | td->num_open_zones++; |
59b07544 BVA |
1150 | z->open = 1; |
1151 | res = true; | |
1152 | ||
1153 | out: | |
1154 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1155 | return res; | |
1156 | } | |
1157 | ||
6463db6c AD |
1158 | /* Anything goes as long as it is not a constant. */ |
1159 | static uint32_t pick_random_zone_idx(const struct fio_file *f, | |
1160 | const struct io_u *io_u) | |
1161 | { | |
1162 | return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size; | |
1163 | } | |
1164 | ||
59b07544 BVA |
1165 | /* |
1166 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
1167 | * that it refers to an open zone. Close an open zone and open a new zone if | |
21c0c884 SK |
1168 | * necessary. The open zone is searched across sequential zones. |
1169 | * This algorithm can only work correctly if all write pointers are | |
59b07544 BVA |
1170 | * a multiple of the fio block size. The caller must neither hold z->mutex |
1171 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1172 | */ | |
379e5f09 BVA |
1173 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
1174 | struct io_u *io_u) | |
59b07544 BVA |
1175 | { |
1176 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 1177 | struct fio_file *f = io_u->file; |
59b07544 BVA |
1178 | struct fio_zone_info *z; |
1179 | unsigned int open_zone_idx = -1; | |
1180 | uint32_t zone_idx, new_zone_idx; | |
1181 | int i; | |
b2da58c4 | 1182 | bool wait_zone_close; |
59b07544 BVA |
1183 | |
1184 | assert(is_valid_offset(f, io_u->offset)); | |
1185 | ||
00ca8df5 | 1186 | if (td->o.max_open_zones || td->o.job_max_open_zones) { |
59b07544 BVA |
1187 | /* |
1188 | * This statement accesses f->zbd_info->open_zones[] on purpose | |
1189 | * without locking. | |
1190 | */ | |
6463db6c | 1191 | zone_idx = f->zbd_info->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 BVA |
1192 | } else { |
1193 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
1194 | } | |
fae3b9a0 AD |
1195 | if (zone_idx < f->min_zone) |
1196 | zone_idx = f->min_zone; | |
1197 | else if (zone_idx >= f->max_zone) | |
1198 | zone_idx = f->max_zone - 1; | |
59b07544 BVA |
1199 | dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", |
1200 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); | |
1201 | ||
1202 | /* | |
1203 | * Since z->mutex is the outer lock and f->zbd_info->mutex the inner | |
1204 | * lock it can happen that the state of the zone with index zone_idx | |
1205 | * has changed after 'z' has been assigned and before f->zbd_info->mutex | |
1206 | * has been obtained. Hence the loop. | |
1207 | */ | |
1208 | for (;;) { | |
6463db6c AD |
1209 | uint32_t tmp_idx; |
1210 | ||
923f7c1e | 1211 | z = get_zone(f, zone_idx); |
14351148 DF |
1212 | if (z->has_wp) |
1213 | zone_lock(td, f, z); | |
59b07544 | 1214 | pthread_mutex_lock(&f->zbd_info->mutex); |
14351148 DF |
1215 | if (z->has_wp) { |
1216 | if (z->cond != ZBD_ZONE_COND_OFFLINE && | |
1217 | td->o.max_open_zones == 0 && td->o.job_max_open_zones == 0) | |
1218 | goto examine_zone; | |
1219 | if (f->zbd_info->num_open_zones == 0) { | |
1220 | dprint(FD_ZBD, "%s(%s): no zones are open\n", | |
1221 | __func__, f->file_name); | |
1222 | goto open_other_zone; | |
1223 | } | |
59b07544 | 1224 | } |
6463db6c AD |
1225 | |
1226 | /* | |
1227 | * List of opened zones is per-device, shared across all threads. | |
1228 | * Start with quasi-random candidate zone. | |
1229 | * Ignore zones which don't belong to thread's offset/size area. | |
1230 | */ | |
1231 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
1a8d510a DF |
1232 | assert(!open_zone_idx || |
1233 | open_zone_idx < f->zbd_info->num_open_zones); | |
6463db6c AD |
1234 | tmp_idx = open_zone_idx; |
1235 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1236 | uint32_t tmpz; | |
1237 | ||
1238 | if (tmp_idx >= f->zbd_info->num_open_zones) | |
1239 | tmp_idx = 0; | |
1240 | tmpz = f->zbd_info->open_zones[tmp_idx]; | |
fae3b9a0 | 1241 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1242 | open_zone_idx = tmp_idx; |
1243 | goto found_candidate_zone; | |
1244 | } | |
1245 | ||
1246 | tmp_idx++; | |
1247 | } | |
1248 | ||
1249 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1250 | __func__, f->file_name); | |
adc6adcb | 1251 | pthread_mutex_unlock(&f->zbd_info->mutex); |
14351148 DF |
1252 | if (z->has_wp) |
1253 | zone_unlock(z); | |
6463db6c AD |
1254 | return NULL; |
1255 | ||
1256 | found_candidate_zone: | |
59b07544 BVA |
1257 | new_zone_idx = f->zbd_info->open_zones[open_zone_idx]; |
1258 | if (new_zone_idx == zone_idx) | |
1259 | break; | |
1260 | zone_idx = new_zone_idx; | |
1261 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
14351148 DF |
1262 | if (z->has_wp) |
1263 | zone_unlock(z); | |
59b07544 BVA |
1264 | } |
1265 | ||
1266 | /* Both z->mutex and f->zbd_info->mutex are held. */ | |
1267 | ||
1268 | examine_zone: | |
236d23a8 | 1269 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) { |
59b07544 BVA |
1270 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1271 | goto out; | |
1272 | } | |
b2da58c4 SK |
1273 | |
1274 | open_other_zone: | |
1275 | /* Check if number of open zones reaches one of limits. */ | |
1276 | wait_zone_close = | |
1277 | f->zbd_info->num_open_zones == f->max_zone - f->min_zone || | |
1278 | (td->o.max_open_zones && | |
1279 | f->zbd_info->num_open_zones == td->o.max_open_zones) || | |
1280 | (td->o.job_max_open_zones && | |
1281 | td->num_open_zones == td->o.job_max_open_zones); | |
1282 | ||
59b07544 BVA |
1283 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1284 | ||
1285 | /* Only z->mutex is held. */ | |
1286 | ||
b2da58c4 SK |
1287 | /* |
1288 | * When number of open zones reaches to one of limits, wait for | |
1289 | * zone close before opening a new zone. | |
1290 | */ | |
1291 | if (wait_zone_close) { | |
1292 | dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n", | |
1293 | __func__, f->file_name); | |
1294 | io_u_quiesce(td); | |
1295 | } | |
1296 | ||
59b07544 BVA |
1297 | /* Zone 'z' is full, so try to open a new zone. */ |
1298 | for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) { | |
1299 | zone_idx++; | |
21c0c884 SK |
1300 | if (z->has_wp) |
1301 | zone_unlock(z); | |
59b07544 | 1302 | z++; |
ee3696bd | 1303 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1304 | /* Wrap-around. */ |
fae3b9a0 | 1305 | zone_idx = f->min_zone; |
923f7c1e | 1306 | z = get_zone(f, zone_idx); |
59b07544 | 1307 | } |
ee3696bd | 1308 | assert(is_valid_offset(f, z->start)); |
21c0c884 SK |
1309 | if (!z->has_wp) |
1310 | continue; | |
fae3b9a0 | 1311 | zone_lock(td, f, z); |
59b07544 BVA |
1312 | if (z->open) |
1313 | continue; | |
954217b9 | 1314 | if (zbd_open_zone(td, f, zone_idx)) |
59b07544 BVA |
1315 | goto out; |
1316 | } | |
1317 | ||
1318 | /* Only z->mutex is held. */ | |
1319 | ||
1320 | /* Check whether the write fits in any of the already opened zones. */ | |
1321 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1322 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1323 | zone_idx = f->zbd_info->open_zones[i]; | |
fae3b9a0 AD |
1324 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1325 | continue; | |
59b07544 | 1326 | pthread_mutex_unlock(&f->zbd_info->mutex); |
4d4c71e6 | 1327 | zone_unlock(z); |
59b07544 | 1328 | |
923f7c1e | 1329 | z = get_zone(f, zone_idx); |
59b07544 | 1330 | |
fae3b9a0 | 1331 | zone_lock(td, f, z); |
236d23a8 | 1332 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) |
59b07544 BVA |
1333 | goto out; |
1334 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1335 | } | |
1336 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
4d4c71e6 | 1337 | zone_unlock(z); |
59b07544 BVA |
1338 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__, |
1339 | f->file_name); | |
1340 | return NULL; | |
1341 | ||
1342 | out: | |
1343 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name, | |
1344 | zone_idx); | |
ee3696bd | 1345 | io_u->offset = z->start; |
21c0c884 | 1346 | assert(z->has_wp); |
8a866de7 | 1347 | assert(z->cond != ZBD_ZONE_COND_OFFLINE); |
59b07544 BVA |
1348 | return z; |
1349 | } | |
1350 | ||
bfbdd35b | 1351 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1352 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1353 | struct io_u *io_u, | |
1354 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1355 | { |
1356 | const struct fio_file *f = io_u->file; | |
1357 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
1358 | ||
122c3539 | 1359 | if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) { |
4d4c71e6 | 1360 | zone_unlock(z); |
59b07544 BVA |
1361 | z = zbd_convert_to_open_zone(td, io_u); |
1362 | assert(z); | |
1363 | } | |
1364 | ||
0cac10b4 | 1365 | if (z->verify_block * min_bs >= z->capacity) { |
a0c84dd4 | 1366 | log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block, |
236d23a8 | 1367 | min_bs, (unsigned long long)z->capacity); |
0cac10b4 DF |
1368 | /* |
1369 | * If the assertion below fails during a test run, adding | |
1370 | * "--experimental_verify=1" to the command line may help. | |
1371 | */ | |
1372 | assert(false); | |
1373 | } | |
1374 | io_u->offset = z->start + z->verify_block * min_bs; | |
1375 | if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1e3d6e03 JA |
1376 | log_err("%s: %llu + %llu >= %llu\n", f->file_name, io_u->offset, |
1377 | io_u->buflen, (unsigned long long) zbd_zone_capacity_end(z)); | |
0cac10b4 DF |
1378 | assert(false); |
1379 | } | |
1380 | z->verify_block += io_u->buflen / min_bs; | |
1381 | ||
59b07544 | 1382 | return z; |
bfbdd35b BVA |
1383 | } |
1384 | ||
1385 | /* | |
21c0c884 SK |
1386 | * Find another zone for which @io_u fits in the readable data in the zone. |
1387 | * Search in zones @zb + 1 .. @zl. For random workload, also search in zones | |
1388 | * @zb - 1 .. @zf. | |
bfbdd35b | 1389 | * |
21c0c884 SK |
1390 | * Either returns NULL or returns a zone pointer. When the zone has write |
1391 | * pointer, hold the mutex for the zone. | |
bfbdd35b BVA |
1392 | */ |
1393 | static struct fio_zone_info * | |
1394 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, | |
1395 | struct fio_zone_info *zb, struct fio_zone_info *zl) | |
1396 | { | |
1397 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 1398 | struct fio_file *f = io_u->file; |
bfbdd35b | 1399 | struct fio_zone_info *z1, *z2; |
923f7c1e | 1400 | const struct fio_zone_info *const zf = get_zone(f, f->min_zone); |
bfbdd35b BVA |
1401 | |
1402 | /* | |
1403 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1404 | * the nearest non-empty zone in case of random I/O. | |
1405 | */ | |
1406 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1407 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1408 | if (z1->has_wp) |
1409 | zone_lock(td, f, z1); | |
ee3696bd | 1410 | if (z1->start + min_bs <= z1->wp) |
bfbdd35b | 1411 | return z1; |
21c0c884 SK |
1412 | if (z1->has_wp) |
1413 | zone_unlock(z1); | |
bfbdd35b BVA |
1414 | } else if (!td_random(td)) { |
1415 | break; | |
1416 | } | |
1417 | if (td_random(td) && z2 >= zf && | |
b7694961 | 1418 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1419 | if (z2->has_wp) |
1420 | zone_lock(td, f, z2); | |
ee3696bd | 1421 | if (z2->start + min_bs <= z2->wp) |
bfbdd35b | 1422 | return z2; |
21c0c884 SK |
1423 | if (z2->has_wp) |
1424 | zone_unlock(z2); | |
bfbdd35b BVA |
1425 | } |
1426 | } | |
1427 | dprint(FD_ZBD, "%s: adjusting random read offset failed\n", | |
1428 | f->file_name); | |
1429 | return NULL; | |
1430 | } | |
1431 | ||
b2da58c4 SK |
1432 | /** |
1433 | * zbd_end_zone_io - update zone status at command completion | |
1434 | * @io_u: I/O unit | |
1435 | * @z: zone info pointer | |
1436 | * | |
1437 | * If the write command made the zone full, close it. | |
1438 | * | |
1439 | * The caller must hold z->mutex. | |
1440 | */ | |
1441 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1442 | struct fio_zone_info *z) | |
1443 | { | |
1444 | const struct fio_file *f = io_u->file; | |
1445 | ||
1446 | if (io_u->ddir == DDIR_WRITE && | |
1447 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1448 | pthread_mutex_lock(&f->zbd_info->mutex); | |
122c3539 | 1449 | zbd_close_zone(td, f, zbd_zone_nr(f, z)); |
b2da58c4 SK |
1450 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1451 | } | |
1452 | } | |
1453 | ||
bfbdd35b | 1454 | /** |
d9ed3e63 | 1455 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1456 | * @io_u: I/O unit |
d9ed3e63 DLM |
1457 | * @success: Whether or not the I/O unit has been queued successfully |
1458 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1459 | * |
d9ed3e63 DLM |
1460 | * For write and trim operations, update the write pointer of the I/O unit |
1461 | * target zone. | |
bfbdd35b | 1462 | */ |
b2da58c4 SK |
1463 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1464 | bool success) | |
bfbdd35b | 1465 | { |
d9ed3e63 DLM |
1466 | const struct fio_file *f = io_u->file; |
1467 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1468 | struct fio_zone_info *z; |
1469 | uint32_t zone_idx; | |
d9ed3e63 | 1470 | uint64_t zone_end; |
bfbdd35b | 1471 | |
5ddf46d0 | 1472 | assert(zbd_info); |
bfbdd35b | 1473 | |
d9ed3e63 | 1474 | zone_idx = zbd_zone_idx(f, io_u->offset); |
bfbdd35b | 1475 | assert(zone_idx < zbd_info->nr_zones); |
923f7c1e | 1476 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1477 | |
43bcbd5b | 1478 | assert(z->has_wp); |
d9ed3e63 | 1479 | |
bfbdd35b BVA |
1480 | if (!success) |
1481 | goto unlock; | |
d9ed3e63 DLM |
1482 | |
1483 | dprint(FD_ZBD, | |
1484 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1485 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1486 | ||
bfbdd35b BVA |
1487 | switch (io_u->ddir) { |
1488 | case DDIR_WRITE: | |
d9ed3e63 | 1489 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1490 | zbd_zone_capacity_end(z)); |
a7c2b6fc BVA |
1491 | pthread_mutex_lock(&zbd_info->mutex); |
1492 | /* | |
1493 | * z->wp > zone_end means that one or more I/O errors | |
1494 | * have occurred. | |
1495 | */ | |
403ea795 | 1496 | if (z->wp <= zone_end) { |
a7c2b6fc | 1497 | zbd_info->sectors_with_data += zone_end - z->wp; |
403ea795 SK |
1498 | zbd_info->wp_sectors_with_data += zone_end - z->wp; |
1499 | } | |
a7c2b6fc | 1500 | pthread_mutex_unlock(&zbd_info->mutex); |
bfbdd35b BVA |
1501 | z->wp = zone_end; |
1502 | break; | |
1503 | case DDIR_TRIM: | |
1504 | assert(z->wp == z->start); | |
1505 | break; | |
1506 | default: | |
1507 | break; | |
1508 | } | |
d9ed3e63 | 1509 | |
b2da58c4 SK |
1510 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1511 | zbd_end_zone_io(td, io_u, z); | |
1512 | ||
bfbdd35b | 1513 | unlock: |
d9ed3e63 DLM |
1514 | if (!success || q != FIO_Q_QUEUED) { |
1515 | /* BUSY or COMPLETED: unlock the zone */ | |
4d4c71e6 | 1516 | zone_unlock(z); |
d9ed3e63 DLM |
1517 | io_u->zbd_put_io = NULL; |
1518 | } | |
1519 | } | |
1520 | ||
1521 | /** | |
1522 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1523 | * @io_u: I/O unit | |
1524 | */ | |
b2da58c4 | 1525 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1526 | { |
1527 | const struct fio_file *f = io_u->file; | |
1528 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1529 | struct fio_zone_info *z; | |
1530 | uint32_t zone_idx; | |
1531 | ||
5ddf46d0 | 1532 | assert(zbd_info); |
615555bb | 1533 | |
d9ed3e63 DLM |
1534 | zone_idx = zbd_zone_idx(f, io_u->offset); |
1535 | assert(zone_idx < zbd_info->nr_zones); | |
923f7c1e | 1536 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1537 | |
43bcbd5b | 1538 | assert(z->has_wp); |
d9ed3e63 DLM |
1539 | |
1540 | dprint(FD_ZBD, | |
1541 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1542 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1543 | ||
b2da58c4 SK |
1544 | zbd_end_zone_io(td, io_u, z); |
1545 | ||
4d4c71e6 | 1546 | zone_unlock(z); |
11afb212 | 1547 | zbd_check_swd(td, f); |
bfbdd35b BVA |
1548 | } |
1549 | ||
9d87c646 DLM |
1550 | /* |
1551 | * Windows and MacOS do not define this. | |
1552 | */ | |
1553 | #ifndef EREMOTEIO | |
1554 | #define EREMOTEIO 121 /* POSIX value */ | |
1555 | #endif | |
1556 | ||
bfbdd35b BVA |
1557 | bool zbd_unaligned_write(int error_code) |
1558 | { | |
1559 | switch (error_code) { | |
1560 | case EIO: | |
1561 | case EREMOTEIO: | |
1562 | return true; | |
1563 | } | |
1564 | return false; | |
1565 | } | |
1566 | ||
4d37720a DLM |
1567 | /** |
1568 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1569 | * @td: FIO thread data. | |
1570 | * @io_u: FIO I/O unit. | |
1571 | * | |
1572 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1573 | * no more IO can be performed in the current zone. | |
1574 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1575 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1576 | * - For write workloads, zoneskip is applied when the zone is full. | |
1577 | * This applies only to read and write operations. | |
1578 | */ | |
1579 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1580 | { | |
1581 | struct fio_file *f = io_u->file; | |
1582 | enum fio_ddir ddir = io_u->ddir; | |
1583 | struct fio_zone_info *z; | |
1584 | uint32_t zone_idx; | |
1585 | ||
1586 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1587 | assert(td->o.zone_size); | |
5ddf46d0 | 1588 | assert(f->zbd_info); |
4d37720a | 1589 | |
236d23a8 | 1590 | zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); |
923f7c1e | 1591 | z = get_zone(f, zone_idx); |
236d23a8 SK |
1592 | |
1593 | /* | |
1594 | * When the zone capacity is smaller than the zone size and the I/O is | |
1595 | * sequential write, skip to zone end if the latest position is at the | |
1596 | * zone capacity limit. | |
1597 | */ | |
1598 | if (z->capacity < f->zbd_info->zone_size && !td_random(td) && | |
1599 | ddir == DDIR_WRITE && | |
1600 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { | |
1601 | dprint(FD_ZBD, | |
1602 | "%s: Jump from zone capacity limit to zone end:" | |
5090d1d0 JA |
1603 | " (%llu -> %llu) for zone %u (%llu)\n", |
1604 | f->file_name, (unsigned long long) f->last_pos[ddir], | |
122c3539 | 1605 | (unsigned long long) zbd_zone_end(z), zone_idx, |
5090d1d0 | 1606 | (unsigned long long) z->capacity); |
236d23a8 SK |
1607 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1608 | f->last_pos[ddir] = zbd_zone_end(z); | |
1609 | } | |
1610 | ||
4d37720a DLM |
1611 | /* |
1612 | * zone_skip is valid only for sequential workloads. | |
1613 | */ | |
1614 | if (td_random(td) || !td->o.zone_skip) | |
1615 | return; | |
1616 | ||
1617 | /* | |
1618 | * It is time to switch to a new zone if: | |
1619 | * - zone_bytes == zone_size bytes have already been accessed | |
1620 | * - The last position reached the end of the current zone. | |
1621 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1622 | * reached the zone write pointer. | |
1623 | */ | |
4d37720a | 1624 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1625 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1626 | (ddir == DDIR_READ && |
1627 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1628 | /* | |
1629 | * Skip zones. | |
1630 | */ | |
1631 | td->zone_bytes = 0; | |
1632 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1633 | ||
1634 | /* | |
1635 | * Wrap from the beginning, if we exceed the file size | |
1636 | */ | |
1637 | if (f->file_offset >= f->real_file_size) | |
1638 | f->file_offset = get_start_offset(td, f); | |
1639 | ||
1640 | f->last_pos[ddir] = f->file_offset; | |
1641 | td->io_skip_bytes += td->o.zone_skip; | |
1642 | } | |
1643 | } | |
1644 | ||
c65057f9 | 1645 | /** |
c7d5e152 | 1646 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1647 | * |
1648 | * @td: FIO thread data. | |
1649 | * @io_u: FIO I/O unit. | |
1650 | * @ddir: I/O direction before adjustment. | |
1651 | * | |
1652 | * Return adjusted I/O direction. | |
1653 | */ | |
1654 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1655 | enum fio_ddir ddir) | |
1656 | { | |
1657 | /* | |
1658 | * In case read direction is chosen for the first random I/O, fio with | |
1659 | * zonemode=zbd stops because no data can be read from zoned block | |
1660 | * devices with all empty zones. Overwrite the first I/O direction as | |
1661 | * write to make sure data to read exists. | |
1662 | */ | |
5ddf46d0 | 1663 | assert(io_u->file->zbd_info); |
731461cc | 1664 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1665 | return ddir; |
1666 | ||
1667 | if (io_u->file->zbd_info->sectors_with_data || | |
1668 | td->o.read_beyond_wp) | |
1669 | return DDIR_READ; | |
1670 | ||
1671 | return DDIR_WRITE; | |
1672 | } | |
1673 | ||
bfbdd35b BVA |
1674 | /** |
1675 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1676 | * @td: FIO thread data. | |
1677 | * @io_u: FIO I/O unit. | |
1678 | * | |
1679 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1680 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1681 | * corresponds to io_u->offset at the end of this function. | |
1682 | */ | |
1683 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1684 | { | |
b7694961 | 1685 | struct fio_file *f = io_u->file; |
bfbdd35b | 1686 | uint32_t zone_idx_b; |
de65f7b7 | 1687 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b BVA |
1688 | uint32_t orig_len = io_u->buflen; |
1689 | uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
1690 | uint64_t new_len; | |
1691 | int64_t range; | |
1692 | ||
5ddf46d0 | 1693 | assert(f->zbd_info); |
adc6adcb | 1694 | assert(min_bs); |
bfbdd35b BVA |
1695 | assert(is_valid_offset(f, io_u->offset)); |
1696 | assert(io_u->buflen); | |
1697 | zone_idx_b = zbd_zone_idx(f, io_u->offset); | |
923f7c1e | 1698 | zb = get_zone(f, zone_idx_b); |
de65f7b7 | 1699 | orig_zb = zb; |
bfbdd35b | 1700 | |
2efcf74b SK |
1701 | if (!zb->has_wp) { |
1702 | /* Accept non-write I/Os for conventional zones. */ | |
1703 | if (io_u->ddir != DDIR_WRITE) | |
1704 | return io_u_accept; | |
1705 | /* | |
1706 | * Make sure that writes to conventional zones | |
1707 | * don't cross over to any sequential zones. | |
1708 | */ | |
1709 | if (!(zb + 1)->has_wp || | |
1710 | io_u->offset + io_u->buflen <= (zb + 1)->start) | |
1711 | return io_u_accept; | |
1712 | ||
1713 | if (io_u->offset + min_bs > (zb + 1)->start) { | |
1714 | dprint(FD_IO, | |
1e3d6e03 JA |
1715 | "%s: off=%llu + min_bs=%u > next zone %llu\n", |
1716 | f->file_name, io_u->offset, | |
1717 | min_bs, (unsigned long long) (zb + 1)->start); | |
2efcf74b SK |
1718 | io_u->offset = zb->start + (zb + 1)->start - io_u->offset; |
1719 | new_len = min(io_u->buflen, (zb + 1)->start - io_u->offset); | |
1720 | } else { | |
1721 | new_len = (zb + 1)->start - io_u->offset; | |
1722 | } | |
1723 | io_u->buflen = new_len / min_bs * min_bs; | |
bfbdd35b | 1724 | return io_u_accept; |
2efcf74b | 1725 | } |
bfbdd35b BVA |
1726 | |
1727 | /* | |
1728 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1729 | * is enabled. | |
1730 | */ | |
b7694961 | 1731 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1732 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1733 | return io_u_accept; | |
1734 | ||
11afb212 | 1735 | zbd_check_swd(td, f); |
615555bb | 1736 | |
fae3b9a0 | 1737 | zone_lock(td, f, zb); |
6f0c6085 | 1738 | |
bfbdd35b BVA |
1739 | switch (io_u->ddir) { |
1740 | case DDIR_READ: | |
0ed7d55e AR |
1741 | if (td->runstate == TD_VERIFYING && td_write(td)) { |
1742 | zb = zbd_replay_write_order(td, io_u, zb); | |
bfbdd35b BVA |
1743 | goto accept; |
1744 | } | |
1745 | /* | |
de65f7b7 DLM |
1746 | * Check that there is enough written data in the zone to do an |
1747 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1748 | * the I/O. | |
bfbdd35b | 1749 | */ |
b7694961 | 1750 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1751 | zb->wp - zb->start : 0; |
de65f7b7 | 1752 | if (range < min_bs || |
ee3696bd | 1753 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
4d4c71e6 | 1754 | zone_unlock(zb); |
923f7c1e | 1755 | zl = get_zone(f, f->max_zone); |
bfbdd35b BVA |
1756 | zb = zbd_find_zone(td, io_u, zb, zl); |
1757 | if (!zb) { | |
1758 | dprint(FD_ZBD, | |
1759 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1760 | f->file_name, io_u->offset, | |
1761 | io_u->buflen); | |
1762 | goto eof; | |
1763 | } | |
de65f7b7 DLM |
1764 | /* |
1765 | * zbd_find_zone() returned a zone with a range of at | |
1766 | * least min_bs. | |
1767 | */ | |
ee3696bd | 1768 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1769 | assert(range >= min_bs); |
1770 | ||
1771 | if (!td_random(td)) | |
ee3696bd | 1772 | io_u->offset = zb->start; |
bfbdd35b | 1773 | } |
de65f7b7 DLM |
1774 | /* |
1775 | * Make sure the I/O is within the zone valid data range while | |
1776 | * maximizing the I/O size and preserving randomness. | |
1777 | */ | |
1778 | if (range <= io_u->buflen) | |
ee3696bd | 1779 | io_u->offset = zb->start; |
de65f7b7 | 1780 | else if (td_random(td)) |
ee3696bd DLM |
1781 | io_u->offset = zb->start + |
1782 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 | 1783 | (range - io_u->buflen)) / min_bs * min_bs; |
43bcbd5b SK |
1784 | /* |
1785 | * When zbd_find_zone() returns a conventional zone, | |
1786 | * we can simply accept the new i/o offset here. | |
1787 | */ | |
1788 | if (!zb->has_wp) | |
1789 | return io_u_accept; | |
de65f7b7 DLM |
1790 | /* |
1791 | * Make sure the I/O does not cross over the zone wp position. | |
1792 | */ | |
1793 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1794 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1795 | new_len = new_len / min_bs * min_bs; |
1796 | if (new_len < io_u->buflen) { | |
1797 | io_u->buflen = new_len; | |
1798 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1799 | orig_len, io_u->buflen); | |
bfbdd35b | 1800 | } |
ee3696bd DLM |
1801 | assert(zb->start <= io_u->offset); |
1802 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
bfbdd35b BVA |
1803 | goto accept; |
1804 | case DDIR_WRITE: | |
1c74aadc DF |
1805 | if (io_u->buflen > f->zbd_info->zone_size) { |
1806 | td_verror(td, EINVAL, "I/O buflen exceeds zone size"); | |
1807 | dprint(FD_IO, | |
1e3d6e03 | 1808 | "%s: I/O buflen %llu exceeds zone size %llu\n", |
1c74aadc | 1809 | f->file_name, io_u->buflen, |
1e3d6e03 | 1810 | (unsigned long long) f->zbd_info->zone_size); |
bfbdd35b | 1811 | goto eof; |
1c74aadc | 1812 | } |
954217b9 | 1813 | if (!zbd_open_zone(td, f, zone_idx_b)) { |
4d4c71e6 | 1814 | zone_unlock(zb); |
59b07544 | 1815 | zb = zbd_convert_to_open_zone(td, io_u); |
1c74aadc DF |
1816 | if (!zb) { |
1817 | dprint(FD_IO, "%s: can't convert to open zone", | |
1818 | f->file_name); | |
59b07544 | 1819 | goto eof; |
1c74aadc | 1820 | } |
122c3539 | 1821 | zone_idx_b = zbd_zone_nr(f, zb); |
59b07544 | 1822 | } |
a7c2b6fc BVA |
1823 | /* Check whether the zone reset threshold has been exceeded */ |
1824 | if (td->o.zrf.u.f) { | |
403ea795 | 1825 | if (f->zbd_info->wp_sectors_with_data >= |
a7c2b6fc BVA |
1826 | f->io_size * td->o.zrt.u.f && |
1827 | zbd_dec_and_reset_write_cnt(td, f)) { | |
1828 | zb->reset_zone = 1; | |
1829 | } | |
1830 | } | |
bfbdd35b BVA |
1831 | /* Reset the zone pointer if necessary */ |
1832 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1833 | assert(td->o.verify == VERIFY_NONE); | |
1834 | /* | |
1835 | * Since previous write requests may have been submitted | |
1836 | * asynchronously and since we will submit the zone | |
1837 | * reset synchronously, wait until previously submitted | |
1838 | * write requests have completed before issuing a | |
1839 | * zone reset. | |
1840 | */ | |
1841 | io_u_quiesce(td); | |
1842 | zb->reset_zone = 0; | |
1843 | if (zbd_reset_zone(td, f, zb) < 0) | |
1844 | goto eof; | |
236d23a8 SK |
1845 | |
1846 | if (zb->capacity < min_bs) { | |
1c74aadc | 1847 | td_verror(td, EINVAL, "ZCAP is less min_bs"); |
236d23a8 SK |
1848 | log_err("zone capacity %llu smaller than minimum block size %d\n", |
1849 | (unsigned long long)zb->capacity, | |
1850 | min_bs); | |
1851 | goto eof; | |
1852 | } | |
bfbdd35b BVA |
1853 | } |
1854 | /* Make writes occur at the write pointer */ | |
1855 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1856 | io_u->offset = zb->wp; |
bfbdd35b | 1857 | if (!is_valid_offset(f, io_u->offset)) { |
1c74aadc DF |
1858 | td_verror(td, EINVAL, "invalid WP value"); |
1859 | dprint(FD_ZBD, "%s: dropped request with offset %llu\n", | |
1860 | f->file_name, io_u->offset); | |
bfbdd35b BVA |
1861 | goto eof; |
1862 | } | |
1863 | /* | |
1864 | * Make sure that the buflen is a multiple of the minimal | |
1865 | * block size. Give up if shrinking would make the request too | |
1866 | * small. | |
1867 | */ | |
1868 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 1869 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
1870 | new_len = new_len / min_bs * min_bs; |
1871 | if (new_len == io_u->buflen) | |
1872 | goto accept; | |
1873 | if (new_len >= min_bs) { | |
1874 | io_u->buflen = new_len; | |
1875 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1876 | orig_len, io_u->buflen); | |
1877 | goto accept; | |
1878 | } | |
1c74aadc DF |
1879 | td_verror(td, EIO, "zone remainder too small"); |
1880 | log_err("zone remainder %lld smaller than min block size %d\n", | |
1881 | (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); | |
bfbdd35b BVA |
1882 | goto eof; |
1883 | case DDIR_TRIM: | |
1884 | /* fall-through */ | |
1885 | case DDIR_SYNC: | |
1886 | case DDIR_DATASYNC: | |
1887 | case DDIR_SYNC_FILE_RANGE: | |
1888 | case DDIR_WAIT: | |
1889 | case DDIR_LAST: | |
1890 | case DDIR_INVAL: | |
1891 | goto accept; | |
1892 | } | |
1893 | ||
1894 | assert(false); | |
1895 | ||
1896 | accept: | |
43bcbd5b | 1897 | assert(zb->has_wp); |
b7694961 | 1898 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
1899 | assert(!io_u->zbd_queue_io); |
1900 | assert(!io_u->zbd_put_io); | |
1901 | io_u->zbd_queue_io = zbd_queue_io; | |
1902 | io_u->zbd_put_io = zbd_put_io; | |
2ef3c1b0 DF |
1903 | /* |
1904 | * Since we return with the zone lock still held, | |
1905 | * add an annotation to let Coverity know that it | |
1906 | * is intentional. | |
1907 | */ | |
1908 | /* coverity[missing_unlock] */ | |
bfbdd35b BVA |
1909 | return io_u_accept; |
1910 | ||
1911 | eof: | |
43bcbd5b | 1912 | if (zb && zb->has_wp) |
4d4c71e6 | 1913 | zone_unlock(zb); |
bfbdd35b BVA |
1914 | return io_u_eof; |
1915 | } | |
fd5d733f BVA |
1916 | |
1917 | /* Return a string with ZBD statistics */ | |
1918 | char *zbd_write_status(const struct thread_stat *ts) | |
1919 | { | |
1920 | char *res; | |
1921 | ||
a0c84dd4 | 1922 | if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0) |
fd5d733f BVA |
1923 | return NULL; |
1924 | return res; | |
1925 | } |