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 | ||
6c5b11d3 DLM |
35 | if (td->io_ops && td->io_ops->get_zoned_model) |
36 | ret = td->io_ops->get_zoned_model(td, f, model); | |
37 | else | |
38 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
39 | if (ret < 0) { |
40 | td_verror(td, errno, "get zoned model failed"); | |
41 | log_err("%s: get zoned model failed (%d).\n", | |
42 | f->file_name, errno); | |
43 | } | |
44 | ||
45 | return ret; | |
46 | } | |
47 | ||
48 | /** | |
49 | * zbd_report_zones - Get zone information | |
50 | * @td: FIO thread data. | |
51 | * @f: FIO file for which to get zone information | |
52 | * @offset: offset from which to report zones | |
53 | * @zones: Array of struct zbd_zone | |
54 | * @nr_zones: Size of @zones array | |
55 | * | |
56 | * Get zone information into @zones starting from the zone at offset @offset | |
57 | * for the device specified by @f. | |
58 | * | |
59 | * Returns the number of zones reported upon success and a negative error code | |
60 | * upon failure. If the zone report is empty, always assume an error (device | |
61 | * problem) and return -EIO. | |
62 | */ | |
63 | int zbd_report_zones(struct thread_data *td, struct fio_file *f, | |
64 | uint64_t offset, struct zbd_zone *zones, | |
65 | unsigned int nr_zones) | |
66 | { | |
67 | int ret; | |
68 | ||
6c5b11d3 DLM |
69 | if (td->io_ops && td->io_ops->report_zones) |
70 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
71 | else | |
72 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
73 | if (ret < 0) { |
74 | td_verror(td, errno, "report zones failed"); | |
75 | log_err("%s: report zones from sector %llu failed (%d).\n", | |
76 | f->file_name, (unsigned long long)offset >> 9, errno); | |
77 | } else if (ret == 0) { | |
78 | td_verror(td, errno, "Empty zone report"); | |
79 | log_err("%s: report zones from sector %llu is empty.\n", | |
80 | f->file_name, (unsigned long long)offset >> 9); | |
81 | ret = -EIO; | |
82 | } | |
83 | ||
84 | return ret; | |
85 | } | |
86 | ||
87 | /** | |
88 | * zbd_reset_wp - reset the write pointer of a range of zones | |
89 | * @td: FIO thread data. | |
90 | * @f: FIO file for which to reset zones | |
91 | * @offset: Starting offset of the first zone to reset | |
92 | * @length: Length of the range of zones to reset | |
93 | * | |
94 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
95 | * Returns 0 upon success and a negative error code upon failure. | |
96 | */ | |
97 | int zbd_reset_wp(struct thread_data *td, struct fio_file *f, | |
98 | uint64_t offset, uint64_t length) | |
99 | { | |
100 | int ret; | |
101 | ||
6c5b11d3 DLM |
102 | if (td->io_ops && td->io_ops->reset_wp) |
103 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
104 | else | |
105 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
106 | if (ret < 0) { |
107 | td_verror(td, errno, "resetting wp failed"); | |
108 | log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n", | |
109 | f->file_name, (unsigned long long)length >> 9, | |
110 | (unsigned long long)offset >> 9, errno); | |
111 | } | |
112 | ||
113 | return ret; | |
114 | } | |
115 | ||
bfbdd35b BVA |
116 | /** |
117 | * zbd_zone_idx - convert an offset into a zone number | |
118 | * @f: file pointer. | |
119 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
120 | * past the disk size then the index of the sentinel is returned. | |
121 | */ | |
122 | static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) | |
123 | { | |
124 | uint32_t zone_idx; | |
125 | ||
cd775e06 | 126 | if (f->zbd_info->zone_size_log2 > 0) |
bfbdd35b BVA |
127 | zone_idx = offset >> f->zbd_info->zone_size_log2; |
128 | else | |
ee3696bd | 129 | zone_idx = offset / f->zbd_info->zone_size; |
bfbdd35b BVA |
130 | |
131 | return min(zone_idx, f->zbd_info->nr_zones); | |
132 | } | |
133 | ||
b7694961 DLM |
134 | /** |
135 | * zbd_zone_swr - Test whether a zone requires sequential writes | |
136 | * @z: zone info pointer. | |
137 | */ | |
138 | static inline bool zbd_zone_swr(struct fio_zone_info *z) | |
139 | { | |
140 | return z->type == ZBD_ZONE_TYPE_SWR; | |
141 | } | |
142 | ||
236d23a8 SK |
143 | /** |
144 | * zbd_zone_end - Return zone end location | |
145 | * @z: zone info pointer. | |
146 | */ | |
147 | static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) | |
148 | { | |
149 | return (z+1)->start; | |
150 | } | |
151 | ||
152 | /** | |
153 | * zbd_zone_capacity_end - Return zone capacity limit end location | |
154 | * @z: zone info pointer. | |
155 | */ | |
156 | static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) | |
157 | { | |
158 | return z->start + z->capacity; | |
159 | } | |
160 | ||
bfbdd35b BVA |
161 | /** |
162 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
163 | * @f: file pointer. | |
164 | * @z: zone info pointer. | |
165 | * @required: minimum number of bytes that must remain in a zone. | |
166 | * | |
167 | * The caller must hold z->mutex. | |
168 | */ | |
169 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
170 | uint64_t required) | |
171 | { | |
172 | assert((required & 511) == 0); | |
173 | ||
b7694961 | 174 | return zbd_zone_swr(z) && |
236d23a8 | 175 | z->wp + required > zbd_zone_capacity_end(z); |
bfbdd35b BVA |
176 | } |
177 | ||
fae3b9a0 | 178 | static void zone_lock(struct thread_data *td, struct fio_file *f, struct fio_zone_info *z) |
1f57803b | 179 | { |
fae3b9a0 AD |
180 | struct zoned_block_device_info *zbd = f->zbd_info; |
181 | uint32_t nz = z - zbd->zone_info; | |
182 | ||
183 | /* A thread should never lock zones outside its working area. */ | |
184 | assert(f->min_zone <= nz && nz < f->max_zone); | |
185 | ||
1f57803b DLM |
186 | /* |
187 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
188 | * is changed or when io_u completes and zbd_put_io() executed. | |
189 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
190 | * other waiting for zone locks when building an io_u batch, first | |
191 | * only trylock the zone. If the zone is already locked by another job, | |
192 | * process the currently queued I/Os so that I/O progress is made and | |
193 | * zones unlocked. | |
194 | */ | |
195 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
196 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
197 | io_u_quiesce(td); | |
198 | pthread_mutex_lock(&z->mutex); | |
199 | } | |
200 | } | |
201 | ||
bfbdd35b BVA |
202 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
203 | { | |
204 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
205 | } | |
206 | ||
923f7c1e DF |
207 | static inline struct fio_zone_info *get_zone(const struct fio_file *f, |
208 | unsigned int zone_nr) | |
209 | { | |
210 | return &f->zbd_info->zone_info[zone_nr]; | |
211 | } | |
212 | ||
bfbdd35b BVA |
213 | /* Verify whether direct I/O is used for all host-managed zoned drives. */ |
214 | static bool zbd_using_direct_io(void) | |
215 | { | |
216 | struct thread_data *td; | |
217 | struct fio_file *f; | |
218 | int i, j; | |
219 | ||
220 | for_each_td(td, i) { | |
221 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) | |
222 | continue; | |
223 | for_each_file(td, f, j) { | |
224 | if (f->zbd_info && | |
b7694961 | 225 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
226 | return false; |
227 | } | |
228 | } | |
229 | ||
230 | return true; | |
231 | } | |
232 | ||
233 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
234 | static bool zbd_is_seq_job(struct fio_file *f) | |
235 | { | |
236 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
237 | ||
238 | assert(f->zbd_info); | |
239 | if (f->io_size == 0) | |
240 | return false; | |
241 | zone_idx_b = zbd_zone_idx(f, f->file_offset); | |
242 | zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); | |
243 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) | |
923f7c1e | 244 | if (zbd_zone_swr(get_zone(f, zone_idx))) |
bfbdd35b BVA |
245 | return true; |
246 | ||
247 | return false; | |
248 | } | |
249 | ||
250 | /* | |
251 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
252 | */ | |
253 | static bool zbd_verify_sizes(void) | |
254 | { | |
255 | const struct fio_zone_info *z; | |
256 | struct thread_data *td; | |
257 | struct fio_file *f; | |
258 | uint64_t new_offset, new_end; | |
259 | uint32_t zone_idx; | |
260 | int i, j; | |
261 | ||
262 | for_each_td(td, i) { | |
263 | for_each_file(td, f, j) { | |
264 | if (!f->zbd_info) | |
265 | continue; | |
266 | if (f->file_offset >= f->real_file_size) | |
267 | continue; | |
268 | if (!zbd_is_seq_job(f)) | |
269 | continue; | |
4d37720a | 270 | |
fc52e449 DLM |
271 | if (!td->o.zone_size) { |
272 | td->o.zone_size = f->zbd_info->zone_size; | |
273 | if (!td->o.zone_size) { | |
274 | log_err("%s: invalid 0 zone size\n", | |
275 | f->file_name); | |
276 | return false; | |
277 | } | |
971d6a22 DLM |
278 | } else if (td->o.zone_size != f->zbd_info->zone_size) { |
279 | log_err("%s: job parameter zonesize %llu does not match disk zone size %llu.\n", | |
280 | f->file_name, (unsigned long long) td->o.zone_size, | |
281 | (unsigned long long) f->zbd_info->zone_size); | |
282 | return false; | |
fc52e449 DLM |
283 | } |
284 | ||
4d37720a DLM |
285 | if (td->o.zone_skip && |
286 | (td->o.zone_skip < td->o.zone_size || | |
287 | td->o.zone_skip % td->o.zone_size)) { | |
288 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
289 | f->file_name, (unsigned long long) td->o.zone_skip, | |
290 | (unsigned long long) td->o.zone_size); | |
291 | return false; | |
292 | } | |
293 | ||
bfbdd35b | 294 | zone_idx = zbd_zone_idx(f, f->file_offset); |
923f7c1e | 295 | z = get_zone(f, zone_idx); |
f09a7773 PL |
296 | if ((f->file_offset != z->start) && |
297 | (td->o.td_ddir != TD_DDIR_READ)) { | |
236d23a8 | 298 | new_offset = zbd_zone_end(z); |
bfbdd35b BVA |
299 | if (new_offset >= f->file_offset + f->io_size) { |
300 | log_info("%s: io_size must be at least one zone\n", | |
301 | f->file_name); | |
302 | return false; | |
303 | } | |
a0c84dd4 JA |
304 | log_info("%s: rounded up offset from %llu to %llu\n", |
305 | f->file_name, (unsigned long long) f->file_offset, | |
306 | (unsigned long long) new_offset); | |
bfbdd35b BVA |
307 | f->io_size -= (new_offset - f->file_offset); |
308 | f->file_offset = new_offset; | |
309 | } | |
310 | zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); | |
923f7c1e | 311 | z = get_zone(f, zone_idx); |
ee3696bd | 312 | new_end = z->start; |
f09a7773 PL |
313 | if ((td->o.td_ddir != TD_DDIR_READ) && |
314 | (f->file_offset + f->io_size != new_end)) { | |
bfbdd35b BVA |
315 | if (new_end <= f->file_offset) { |
316 | log_info("%s: io_size must be at least one zone\n", | |
317 | f->file_name); | |
318 | return false; | |
319 | } | |
a0c84dd4 JA |
320 | log_info("%s: rounded down io_size from %llu to %llu\n", |
321 | f->file_name, (unsigned long long) f->io_size, | |
322 | (unsigned long long) new_end - f->file_offset); | |
bfbdd35b BVA |
323 | f->io_size = new_end - f->file_offset; |
324 | } | |
fae3b9a0 AD |
325 | |
326 | f->min_zone = zbd_zone_idx(f, f->file_offset); | |
327 | f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size); | |
5c79c32f | 328 | assert(f->min_zone < f->max_zone); |
bfbdd35b BVA |
329 | } |
330 | } | |
331 | ||
332 | return true; | |
333 | } | |
334 | ||
335 | static bool zbd_verify_bs(void) | |
336 | { | |
337 | struct thread_data *td; | |
338 | struct fio_file *f; | |
339 | uint32_t zone_size; | |
340 | int i, j, k; | |
341 | ||
342 | for_each_td(td, i) { | |
343 | for_each_file(td, f, j) { | |
344 | if (!f->zbd_info) | |
345 | continue; | |
346 | zone_size = f->zbd_info->zone_size; | |
59f94d26 | 347 | for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) { |
bfbdd35b | 348 | if (td->o.verify != VERIFY_NONE && |
ee3696bd | 349 | zone_size % td->o.bs[k] != 0) { |
bfbdd35b BVA |
350 | log_info("%s: block size %llu is not a divisor of the zone size %d\n", |
351 | f->file_name, td->o.bs[k], | |
ee3696bd | 352 | zone_size); |
bfbdd35b BVA |
353 | return false; |
354 | } | |
355 | } | |
356 | } | |
357 | } | |
358 | return true; | |
359 | } | |
360 | ||
bfbdd35b BVA |
361 | static int ilog2(uint64_t i) |
362 | { | |
363 | int log = -1; | |
364 | ||
365 | while (i) { | |
366 | i >>= 1; | |
367 | log++; | |
368 | } | |
369 | return log; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
374 | * allows to execute a ZBD workload against a non-ZBD device. | |
375 | */ | |
376 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
377 | { | |
378 | uint32_t nr_zones; | |
379 | struct fio_zone_info *p; | |
a4b7f12b | 380 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 381 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 382 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
383 | int i; |
384 | ||
a4b7f12b DLM |
385 | if (zone_size == 0) { |
386 | log_err("%s: Specifying the zone size is mandatory for regular block devices with --zonemode=zbd\n\n", | |
387 | f->file_name); | |
388 | return 1; | |
389 | } | |
390 | ||
391 | if (zone_size < 512) { | |
392 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
393 | f->file_name); | |
394 | return 1; | |
395 | } | |
396 | ||
b8dd9750 HH |
397 | if (zone_capacity == 0) |
398 | zone_capacity = zone_size; | |
399 | ||
400 | if (zone_capacity > zone_size) { | |
401 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
402 | f->file_name, (unsigned long long) td->o.zone_capacity, | |
403 | (unsigned long long) td->o.zone_size); | |
404 | return 1; | |
405 | } | |
406 | ||
ee3696bd | 407 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
408 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
409 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
410 | if (!zbd_info) | |
411 | return -ENOMEM; | |
412 | ||
44ec32cb | 413 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
414 | zbd_info->refcount = 1; |
415 | p = &zbd_info->zone_info[0]; | |
416 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
417 | mutex_init_pshared_with_type(&p->mutex, |
418 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 419 | p->start = i * zone_size; |
b14651a2 | 420 | p->wp = p->start; |
b7694961 DLM |
421 | p->type = ZBD_ZONE_TYPE_SWR; |
422 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 423 | p->capacity = zone_capacity; |
bfbdd35b BVA |
424 | } |
425 | /* a sentinel */ | |
426 | p->start = nr_zones * zone_size; | |
427 | ||
428 | f->zbd_info = zbd_info; | |
429 | f->zbd_info->zone_size = zone_size; | |
430 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 431 | ilog2(zone_size) : 0; |
bfbdd35b | 432 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
433 | return 0; |
434 | } | |
435 | ||
436 | /* | |
b7694961 DLM |
437 | * Maximum number of zones to report in one operation. |
438 | */ | |
439 | #define ZBD_REPORT_MAX_ZONES 8192U | |
440 | ||
441 | /* | |
442 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
443 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
444 | */ |
445 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
446 | { | |
b7694961 DLM |
447 | int nr_zones, nrz; |
448 | struct zbd_zone *zones, *z; | |
bfbdd35b | 449 | struct fio_zone_info *p; |
b7694961 | 450 | uint64_t zone_size, offset; |
bfbdd35b | 451 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 452 | int i, j, ret = -ENOMEM; |
bfbdd35b | 453 | |
b7694961 DLM |
454 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
455 | if (!zones) | |
bfbdd35b BVA |
456 | goto out; |
457 | ||
b7694961 DLM |
458 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
459 | if (nrz < 0) { | |
460 | ret = nrz; | |
461 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
462 | f->file_name, -ret); | |
463 | goto out; | |
bfbdd35b BVA |
464 | } |
465 | ||
b7694961 | 466 | zone_size = zones[0].len; |
ee3696bd | 467 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
468 | |
469 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
470 | td->o.zone_size = zone_size; |
471 | } else if (td->o.zone_size != zone_size) { | |
4580ff92 DLM |
472 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n", |
473 | f->file_name, (unsigned long long) td->o.zone_size, | |
a0c84dd4 | 474 | (unsigned long long) zone_size); |
bfbdd35b | 475 | ret = -EINVAL; |
b7694961 | 476 | goto out; |
bfbdd35b BVA |
477 | } |
478 | ||
a0c84dd4 JA |
479 | dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name, |
480 | nr_zones, (unsigned long long) zone_size / 1024); | |
bfbdd35b BVA |
481 | |
482 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
483 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 484 | if (!zbd_info) |
b7694961 | 485 | goto out; |
44ec32cb | 486 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
487 | zbd_info->refcount = 1; |
488 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
489 | for (offset = 0, j = 0; j < nr_zones;) { |
490 | z = &zones[0]; | |
491 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
492 | mutex_init_pshared_with_type(&p->mutex, |
493 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 494 | p->start = z->start; |
236d23a8 | 495 | p->capacity = z->capacity; |
bfbdd35b | 496 | switch (z->cond) { |
b7694961 DLM |
497 | case ZBD_ZONE_COND_NOT_WP: |
498 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 499 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
500 | break; |
501 | default: | |
502 | assert(z->start <= z->wp); | |
b7694961 DLM |
503 | assert(z->wp <= z->start + zone_size); |
504 | p->wp = z->wp; | |
bfbdd35b BVA |
505 | break; |
506 | } | |
507 | p->type = z->type; | |
508 | p->cond = z->cond; | |
509 | if (j > 0 && p->start != p[-1].start + zone_size) { | |
510 | log_info("%s: invalid zone data\n", | |
511 | f->file_name); | |
512 | ret = -EINVAL; | |
b7694961 | 513 | goto out; |
bfbdd35b BVA |
514 | } |
515 | } | |
516 | z--; | |
b7694961 | 517 | offset = z->start + z->len; |
bfbdd35b BVA |
518 | if (j >= nr_zones) |
519 | break; | |
b7694961 DLM |
520 | nrz = zbd_report_zones(td, f, offset, |
521 | zones, ZBD_REPORT_MAX_ZONES); | |
522 | if (nrz < 0) { | |
523 | ret = nrz; | |
524 | log_info("fio: report zones (offset %llu) failed for %s (%d).\n", | |
525 | (unsigned long long)offset, | |
526 | f->file_name, -ret); | |
527 | goto out; | |
bfbdd35b BVA |
528 | } |
529 | } | |
b7694961 | 530 | |
bfbdd35b | 531 | /* a sentinel */ |
b7694961 | 532 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
533 | |
534 | f->zbd_info = zbd_info; | |
535 | f->zbd_info->zone_size = zone_size; | |
536 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 537 | ilog2(zone_size) : 0; |
bfbdd35b BVA |
538 | f->zbd_info->nr_zones = nr_zones; |
539 | zbd_info = NULL; | |
540 | ret = 0; | |
541 | ||
bfbdd35b | 542 | out: |
b7694961 DLM |
543 | sfree(zbd_info); |
544 | free(zones); | |
bfbdd35b BVA |
545 | return ret; |
546 | } | |
547 | ||
548 | /* | |
549 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
550 | * | |
551 | * Returns 0 upon success and a negative error code upon failure. | |
552 | */ | |
379e5f09 | 553 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 554 | { |
b7694961 DLM |
555 | enum zbd_zoned_model zbd_model; |
556 | int ret; | |
bfbdd35b BVA |
557 | |
558 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
559 | ||
b7694961 DLM |
560 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
561 | if (ret) | |
562 | return ret; | |
563 | ||
bfbdd35b | 564 | switch (zbd_model) { |
b7694961 DLM |
565 | case ZBD_IGNORE: |
566 | return 0; | |
567 | case ZBD_HOST_AWARE: | |
568 | case ZBD_HOST_MANAGED: | |
bfbdd35b BVA |
569 | ret = parse_zone_info(td, f); |
570 | break; | |
b7694961 | 571 | case ZBD_NONE: |
bfbdd35b BVA |
572 | ret = init_zone_info(td, f); |
573 | break; | |
b7694961 DLM |
574 | default: |
575 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
576 | log_err("Unsupported zoned model\n"); | |
577 | return -EINVAL; | |
bfbdd35b | 578 | } |
b7694961 | 579 | |
219c662d | 580 | if (ret == 0) { |
bfbdd35b | 581 | f->zbd_info->model = zbd_model; |
219c662d AD |
582 | f->zbd_info->max_open_zones = td->o.max_open_zones; |
583 | } | |
bfbdd35b BVA |
584 | return ret; |
585 | } | |
586 | ||
587 | void zbd_free_zone_info(struct fio_file *f) | |
588 | { | |
589 | uint32_t refcount; | |
590 | ||
3c1dc34c | 591 | assert(f->zbd_info); |
bfbdd35b BVA |
592 | |
593 | pthread_mutex_lock(&f->zbd_info->mutex); | |
594 | refcount = --f->zbd_info->refcount; | |
595 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
596 | ||
597 | assert((int32_t)refcount >= 0); | |
598 | if (refcount == 0) | |
599 | sfree(f->zbd_info); | |
600 | f->zbd_info = NULL; | |
601 | } | |
602 | ||
603 | /* | |
604 | * Initialize f->zbd_info. | |
605 | * | |
606 | * Returns 0 upon success and a negative error code upon failure. | |
607 | * | |
608 | * Note: this function can only work correctly if it is called before the first | |
609 | * fio fork() call. | |
610 | */ | |
611 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
612 | { | |
613 | struct thread_data *td2; | |
614 | struct fio_file *f2; | |
615 | int i, j, ret; | |
616 | ||
617 | for_each_td(td2, i) { | |
618 | for_each_file(td2, f2, j) { | |
619 | if (td2 == td && f2 == file) | |
620 | continue; | |
621 | if (!f2->zbd_info || | |
622 | strcmp(f2->file_name, file->file_name) != 0) | |
623 | continue; | |
624 | file->zbd_info = f2->zbd_info; | |
625 | file->zbd_info->refcount++; | |
626 | return 0; | |
627 | } | |
628 | } | |
629 | ||
630 | ret = zbd_create_zone_info(td, file); | |
631 | if (ret < 0) | |
c5837eec | 632 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
bfbdd35b BVA |
633 | return ret; |
634 | } | |
635 | ||
954217b9 SK |
636 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
637 | uint32_t zone_idx); | |
638 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
639 | struct fio_zone_info *z); | |
640 | ||
3c1dc34c | 641 | int zbd_setup_files(struct thread_data *td) |
bfbdd35b BVA |
642 | { |
643 | struct fio_file *f; | |
644 | int i; | |
645 | ||
646 | for_each_file(td, f, i) { | |
a4b7f12b | 647 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 648 | return 1; |
bfbdd35b BVA |
649 | } |
650 | ||
651 | if (!zbd_using_direct_io()) { | |
652 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
653 | return 1; | |
654 | } | |
655 | ||
656 | if (!zbd_verify_sizes()) | |
657 | return 1; | |
658 | ||
659 | if (!zbd_verify_bs()) | |
660 | return 1; | |
661 | ||
219c662d AD |
662 | for_each_file(td, f, i) { |
663 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
664 | struct fio_zone_info *z; |
665 | int zi; | |
219c662d AD |
666 | |
667 | if (!zbd) | |
668 | continue; | |
669 | ||
670 | zbd->max_open_zones = zbd->max_open_zones ?: ZBD_MAX_OPEN_ZONES; | |
671 | ||
672 | if (td->o.max_open_zones > 0 && | |
673 | zbd->max_open_zones != td->o.max_open_zones) { | |
674 | log_err("Different 'max_open_zones' values\n"); | |
675 | return 1; | |
676 | } | |
677 | if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) { | |
678 | log_err("'max_open_zones' value is limited by %u\n", ZBD_MAX_OPEN_ZONES); | |
679 | return 1; | |
680 | } | |
954217b9 SK |
681 | |
682 | for (zi = f->min_zone; zi < f->max_zone; zi++) { | |
683 | z = &zbd->zone_info[zi]; | |
684 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
685 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
686 | continue; | |
687 | if (zbd_open_zone(td, f, zi)) | |
688 | continue; | |
689 | /* | |
690 | * If the number of open zones exceeds specified limits, | |
691 | * reset all extra open zones. | |
692 | */ | |
693 | if (zbd_reset_zone(td, f, z) < 0) { | |
694 | log_err("Failed to reest zone %d\n", zi); | |
695 | return 1; | |
696 | } | |
697 | } | |
219c662d AD |
698 | } |
699 | ||
bfbdd35b BVA |
700 | return 0; |
701 | } | |
702 | ||
122c3539 DF |
703 | static inline unsigned int zbd_zone_nr(const struct fio_file *f, |
704 | struct fio_zone_info *zone) | |
ceb35a11 | 705 | { |
122c3539 | 706 | return zone - f->zbd_info->zone_info; |
ceb35a11 DF |
707 | } |
708 | ||
bfbdd35b | 709 | /** |
ceb35a11 | 710 | * zbd_reset_zone - reset the write pointer of a single zone |
bfbdd35b | 711 | * @td: FIO thread data. |
ceb35a11 DF |
712 | * @f: FIO file associated with the disk for which to reset a write pointer. |
713 | * @z: Zone to reset. | |
bfbdd35b BVA |
714 | * |
715 | * Returns 0 upon success and a negative error code upon failure. | |
ceb35a11 DF |
716 | * |
717 | * The caller must hold z->mutex. | |
bfbdd35b | 718 | */ |
ceb35a11 DF |
719 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, |
720 | struct fio_zone_info *z) | |
bfbdd35b | 721 | { |
ceb35a11 DF |
722 | uint64_t offset = z->start; |
723 | uint64_t length = (z+1)->start - offset; | |
bfbdd35b BVA |
724 | int ret = 0; |
725 | ||
071c9d29 SK |
726 | if (z->wp == z->start) |
727 | return 0; | |
728 | ||
ee3696bd | 729 | assert(is_valid_offset(f, offset + length - 1)); |
b7694961 | 730 | |
ceb35a11 | 731 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name, |
122c3539 | 732 | zbd_zone_nr(f, z)); |
bfbdd35b | 733 | switch (f->zbd_info->model) { |
b7694961 DLM |
734 | case ZBD_HOST_AWARE: |
735 | case ZBD_HOST_MANAGED: | |
736 | ret = zbd_reset_wp(td, f, offset, length); | |
737 | if (ret < 0) | |
bfbdd35b | 738 | return ret; |
bfbdd35b | 739 | break; |
b7694961 | 740 | default: |
bfbdd35b BVA |
741 | break; |
742 | } | |
743 | ||
ceb35a11 DF |
744 | pthread_mutex_lock(&f->zbd_info->mutex); |
745 | f->zbd_info->sectors_with_data -= z->wp - z->start; | |
746 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
747 | z->wp = z->start; | |
748 | z->verify_block = 0; | |
bfbdd35b | 749 | |
ceb35a11 | 750 | td->ts.nr_zone_resets++; |
fd5d733f | 751 | |
bfbdd35b BVA |
752 | return ret; |
753 | } | |
754 | ||
25dc6606 AD |
755 | /* The caller must hold f->zbd_info->mutex */ |
756 | static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, | |
b2da58c4 | 757 | unsigned int zone_idx) |
25dc6606 | 758 | { |
b2da58c4 | 759 | uint32_t open_zone_idx = 0; |
25dc6606 | 760 | |
b2da58c4 SK |
761 | for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) { |
762 | if (f->zbd_info->open_zones[open_zone_idx] == zone_idx) | |
763 | break; | |
764 | } | |
765 | if (open_zone_idx == f->zbd_info->num_open_zones) { | |
766 | dprint(FD_ZBD, "%s: zone %d is not open\n", | |
767 | f->file_name, zone_idx); | |
768 | return; | |
769 | } | |
770 | ||
771 | dprint(FD_ZBD, "%s: closing zone %d\n", f->file_name, zone_idx); | |
25dc6606 AD |
772 | memmove(f->zbd_info->open_zones + open_zone_idx, |
773 | f->zbd_info->open_zones + open_zone_idx + 1, | |
774 | (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) * | |
775 | sizeof(f->zbd_info->open_zones[0])); | |
776 | f->zbd_info->num_open_zones--; | |
219c662d | 777 | td->num_open_zones--; |
923f7c1e | 778 | get_zone(f, zone_idx)->open = 0; |
25dc6606 AD |
779 | } |
780 | ||
bfbdd35b BVA |
781 | /* |
782 | * Reset a range of zones. Returns 0 upon success and 1 upon failure. | |
783 | * @td: fio thread data. | |
784 | * @f: fio file for which to reset zones | |
785 | * @zb: first zone to reset. | |
786 | * @ze: first zone not to reset. | |
787 | * @all_zones: whether to reset all zones or only those zones for which the | |
788 | * write pointer is not a multiple of td->o.min_bs[DDIR_WRITE]. | |
789 | */ | |
790 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
791 | struct fio_zone_info *const zb, | |
792 | struct fio_zone_info *const ze, bool all_zones) | |
793 | { | |
4803b841 | 794 | struct fio_zone_info *z; |
ee3696bd | 795 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; |
bfbdd35b BVA |
796 | bool reset_wp; |
797 | int res = 0; | |
798 | ||
adc6adcb DLM |
799 | assert(min_bs); |
800 | ||
a0c84dd4 | 801 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name, |
122c3539 | 802 | zbd_zone_nr(f, zb), zbd_zone_nr(f, ze)); |
bfbdd35b | 803 | for (z = zb; z < ze; z++) { |
122c3539 | 804 | uint32_t nz = zbd_zone_nr(f, z); |
25dc6606 | 805 | |
b7694961 | 806 | if (!zbd_zone_swr(z)) |
1f57803b | 807 | continue; |
fae3b9a0 | 808 | zone_lock(td, f, z); |
25dc6606 | 809 | if (all_zones) { |
25dc6606 | 810 | pthread_mutex_lock(&f->zbd_info->mutex); |
b2da58c4 | 811 | zbd_close_zone(td, f, nz); |
25dc6606 AD |
812 | pthread_mutex_unlock(&f->zbd_info->mutex); |
813 | ||
814 | reset_wp = z->wp != z->start; | |
815 | } else { | |
767d1372 | 816 | reset_wp = z->wp % min_bs != 0; |
25dc6606 | 817 | } |
1f57803b DLM |
818 | if (reset_wp) { |
819 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
122c3539 | 820 | f->file_name, zbd_zone_nr(f, z)); |
1f57803b DLM |
821 | if (zbd_reset_zone(td, f, z) < 0) |
822 | res = 1; | |
bfbdd35b | 823 | } |
bfbdd35b | 824 | pthread_mutex_unlock(&z->mutex); |
4803b841 | 825 | } |
bfbdd35b BVA |
826 | |
827 | return res; | |
828 | } | |
829 | ||
a7c2b6fc BVA |
830 | /* |
831 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
832 | * zone reset. | |
833 | */ | |
1bb1bcad AD |
834 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
835 | const struct fio_file *f) | |
a7c2b6fc BVA |
836 | { |
837 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
838 | ||
a7c2b6fc BVA |
839 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
840 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
841 | } |
842 | ||
843 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
844 | const struct fio_file *f) | |
845 | { | |
846 | pthread_mutex_lock(&f->zbd_info->mutex); | |
847 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
848 | pthread_mutex_unlock(&f->zbd_info->mutex); |
849 | } | |
850 | ||
851 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
852 | const struct fio_file *f) | |
853 | { | |
854 | uint32_t write_cnt = 0; | |
855 | ||
856 | pthread_mutex_lock(&f->zbd_info->mutex); | |
857 | assert(f->zbd_info->write_cnt); | |
858 | if (f->zbd_info->write_cnt) | |
859 | write_cnt = --f->zbd_info->write_cnt; | |
860 | if (write_cnt == 0) | |
1bb1bcad | 861 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
862 | pthread_mutex_unlock(&f->zbd_info->mutex); |
863 | ||
864 | return write_cnt == 0; | |
865 | } | |
866 | ||
91d25131 BVA |
867 | enum swd_action { |
868 | CHECK_SWD, | |
869 | SET_SWD, | |
870 | }; | |
871 | ||
872 | /* Calculate the number of sectors with data (swd) and perform action 'a' */ | |
873 | static uint64_t zbd_process_swd(const struct fio_file *f, enum swd_action a) | |
615555bb | 874 | { |
615555bb BVA |
875 | struct fio_zone_info *zb, *ze, *z; |
876 | uint64_t swd = 0; | |
877 | ||
923f7c1e DF |
878 | zb = get_zone(f, f->min_zone); |
879 | ze = get_zone(f, f->max_zone); | |
615555bb BVA |
880 | for (z = zb; z < ze; z++) { |
881 | pthread_mutex_lock(&z->mutex); | |
882 | swd += z->wp - z->start; | |
883 | } | |
884 | pthread_mutex_lock(&f->zbd_info->mutex); | |
91d25131 BVA |
885 | switch (a) { |
886 | case CHECK_SWD: | |
887 | assert(f->zbd_info->sectors_with_data == swd); | |
888 | break; | |
889 | case SET_SWD: | |
890 | f->zbd_info->sectors_with_data = swd; | |
891 | break; | |
892 | } | |
615555bb BVA |
893 | pthread_mutex_unlock(&f->zbd_info->mutex); |
894 | for (z = zb; z < ze; z++) | |
895 | pthread_mutex_unlock(&z->mutex); | |
91d25131 BVA |
896 | |
897 | return swd; | |
898 | } | |
899 | ||
900 | /* | |
901 | * The swd check is useful for debugging but takes too much time to leave | |
902 | * it enabled all the time. Hence it is disabled by default. | |
903 | */ | |
904 | static const bool enable_check_swd = false; | |
905 | ||
906 | /* Check whether the value of zbd_info.sectors_with_data is correct. */ | |
907 | static void zbd_check_swd(const struct fio_file *f) | |
908 | { | |
909 | if (!enable_check_swd) | |
910 | return; | |
911 | ||
912 | zbd_process_swd(f, CHECK_SWD); | |
913 | } | |
914 | ||
915 | static void zbd_init_swd(struct fio_file *f) | |
916 | { | |
917 | uint64_t swd; | |
918 | ||
409a4f29 NA |
919 | if (!enable_check_swd) |
920 | return; | |
921 | ||
91d25131 BVA |
922 | swd = zbd_process_swd(f, SET_SWD); |
923 | dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name, | |
924 | swd); | |
615555bb BVA |
925 | } |
926 | ||
bfbdd35b BVA |
927 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
928 | { | |
91d25131 | 929 | struct fio_zone_info *zb, *ze; |
bfbdd35b | 930 | |
767d1372 | 931 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
932 | return; |
933 | ||
923f7c1e DF |
934 | zb = get_zone(f, f->min_zone); |
935 | ze = get_zone(f, f->max_zone); | |
91d25131 | 936 | zbd_init_swd(f); |
bfbdd35b BVA |
937 | /* |
938 | * If data verification is enabled reset the affected zones before | |
939 | * writing any data to avoid that a zone reset has to be issued while | |
940 | * writing data, which causes data loss. | |
941 | */ | |
942 | zbd_reset_zones(td, f, zb, ze, td->o.verify != VERIFY_NONE && | |
bfbdd35b | 943 | td->runstate != TD_VERIFYING); |
a7c2b6fc | 944 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
945 | } |
946 | ||
59b07544 BVA |
947 | /* The caller must hold f->zbd_info->mutex. */ |
948 | static bool is_zone_open(const struct thread_data *td, const struct fio_file *f, | |
949 | unsigned int zone_idx) | |
950 | { | |
951 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
952 | int i; | |
953 | ||
219c662d AD |
954 | assert(td->o.job_max_open_zones == 0 || td->num_open_zones <= td->o.job_max_open_zones); |
955 | assert(td->o.job_max_open_zones <= zbdi->max_open_zones); | |
956 | assert(zbdi->num_open_zones <= zbdi->max_open_zones); | |
59b07544 BVA |
957 | |
958 | for (i = 0; i < zbdi->num_open_zones; i++) | |
959 | if (zbdi->open_zones[i] == zone_idx) | |
960 | return true; | |
961 | ||
962 | return false; | |
963 | } | |
964 | ||
965 | /* | |
966 | * Open a ZBD zone if it was not yet open. Returns true if either the zone was | |
967 | * already open or if opening a new zone is allowed. Returns false if the zone | |
968 | * was not yet open and opening a new zone would cause the zone limit to be | |
969 | * exceeded. | |
970 | */ | |
954217b9 | 971 | static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f, |
59b07544 BVA |
972 | uint32_t zone_idx) |
973 | { | |
974 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
923f7c1e | 975 | struct fio_zone_info *z = get_zone(f, zone_idx); |
59b07544 BVA |
976 | bool res = true; |
977 | ||
b7694961 | 978 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
59b07544 BVA |
979 | return false; |
980 | ||
981 | /* | |
982 | * Skip full zones with data verification enabled because resetting a | |
983 | * zone causes data loss and hence causes verification to fail. | |
984 | */ | |
985 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) | |
986 | return false; | |
987 | ||
59b07544 | 988 | pthread_mutex_lock(&f->zbd_info->mutex); |
b2da58c4 SK |
989 | if (is_zone_open(td, f, zone_idx)) { |
990 | /* | |
991 | * If the zone is already open and going to be full by writes | |
992 | * in-flight, handle it as a full zone instead of an open zone. | |
993 | */ | |
994 | if (z->wp >= zbd_zone_capacity_end(z)) | |
995 | res = false; | |
59b07544 | 996 | goto out; |
b2da58c4 | 997 | } |
59b07544 | 998 | res = false; |
219c662d AD |
999 | /* Zero means no limit */ |
1000 | if (td->o.job_max_open_zones > 0 && | |
1001 | td->num_open_zones >= td->o.job_max_open_zones) | |
1002 | goto out; | |
1003 | if (f->zbd_info->num_open_zones >= f->zbd_info->max_open_zones) | |
59b07544 BVA |
1004 | goto out; |
1005 | dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx); | |
1006 | f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx; | |
219c662d | 1007 | td->num_open_zones++; |
59b07544 BVA |
1008 | z->open = 1; |
1009 | res = true; | |
1010 | ||
1011 | out: | |
1012 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1013 | return res; | |
1014 | } | |
1015 | ||
6463db6c AD |
1016 | /* Anything goes as long as it is not a constant. */ |
1017 | static uint32_t pick_random_zone_idx(const struct fio_file *f, | |
1018 | const struct io_u *io_u) | |
1019 | { | |
1020 | return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size; | |
1021 | } | |
1022 | ||
59b07544 BVA |
1023 | /* |
1024 | * Modify the offset of an I/O unit that does not refer to an open zone such | |
1025 | * that it refers to an open zone. Close an open zone and open a new zone if | |
1026 | * necessary. This algorithm can only work correctly if all write pointers are | |
1027 | * a multiple of the fio block size. The caller must neither hold z->mutex | |
1028 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1029 | */ | |
379e5f09 BVA |
1030 | static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, |
1031 | struct io_u *io_u) | |
59b07544 BVA |
1032 | { |
1033 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 1034 | struct fio_file *f = io_u->file; |
59b07544 BVA |
1035 | struct fio_zone_info *z; |
1036 | unsigned int open_zone_idx = -1; | |
1037 | uint32_t zone_idx, new_zone_idx; | |
1038 | int i; | |
b2da58c4 | 1039 | bool wait_zone_close; |
59b07544 BVA |
1040 | |
1041 | assert(is_valid_offset(f, io_u->offset)); | |
1042 | ||
00ca8df5 | 1043 | if (td->o.max_open_zones || td->o.job_max_open_zones) { |
59b07544 BVA |
1044 | /* |
1045 | * This statement accesses f->zbd_info->open_zones[] on purpose | |
1046 | * without locking. | |
1047 | */ | |
6463db6c | 1048 | zone_idx = f->zbd_info->open_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 BVA |
1049 | } else { |
1050 | zone_idx = zbd_zone_idx(f, io_u->offset); | |
1051 | } | |
fae3b9a0 AD |
1052 | if (zone_idx < f->min_zone) |
1053 | zone_idx = f->min_zone; | |
1054 | else if (zone_idx >= f->max_zone) | |
1055 | zone_idx = f->max_zone - 1; | |
59b07544 BVA |
1056 | dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", |
1057 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); | |
1058 | ||
1059 | /* | |
1060 | * Since z->mutex is the outer lock and f->zbd_info->mutex the inner | |
1061 | * lock it can happen that the state of the zone with index zone_idx | |
1062 | * has changed after 'z' has been assigned and before f->zbd_info->mutex | |
1063 | * has been obtained. Hence the loop. | |
1064 | */ | |
1065 | for (;;) { | |
6463db6c AD |
1066 | uint32_t tmp_idx; |
1067 | ||
923f7c1e | 1068 | z = get_zone(f, zone_idx); |
59b07544 | 1069 | |
fae3b9a0 | 1070 | zone_lock(td, f, z); |
59b07544 | 1071 | pthread_mutex_lock(&f->zbd_info->mutex); |
00ca8df5 | 1072 | if (td->o.max_open_zones == 0 && td->o.job_max_open_zones == 0) |
59b07544 BVA |
1073 | goto examine_zone; |
1074 | if (f->zbd_info->num_open_zones == 0) { | |
59b07544 BVA |
1075 | dprint(FD_ZBD, "%s(%s): no zones are open\n", |
1076 | __func__, f->file_name); | |
b2da58c4 | 1077 | goto open_other_zone; |
59b07544 | 1078 | } |
6463db6c AD |
1079 | |
1080 | /* | |
1081 | * List of opened zones is per-device, shared across all threads. | |
1082 | * Start with quasi-random candidate zone. | |
1083 | * Ignore zones which don't belong to thread's offset/size area. | |
1084 | */ | |
1085 | open_zone_idx = pick_random_zone_idx(f, io_u); | |
59b07544 | 1086 | assert(open_zone_idx < f->zbd_info->num_open_zones); |
6463db6c AD |
1087 | tmp_idx = open_zone_idx; |
1088 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1089 | uint32_t tmpz; | |
1090 | ||
1091 | if (tmp_idx >= f->zbd_info->num_open_zones) | |
1092 | tmp_idx = 0; | |
1093 | tmpz = f->zbd_info->open_zones[tmp_idx]; | |
fae3b9a0 | 1094 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
6463db6c AD |
1095 | open_zone_idx = tmp_idx; |
1096 | goto found_candidate_zone; | |
1097 | } | |
1098 | ||
1099 | tmp_idx++; | |
1100 | } | |
1101 | ||
1102 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1103 | __func__, f->file_name); | |
adc6adcb DLM |
1104 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1105 | pthread_mutex_unlock(&z->mutex); | |
6463db6c AD |
1106 | return NULL; |
1107 | ||
1108 | found_candidate_zone: | |
59b07544 BVA |
1109 | new_zone_idx = f->zbd_info->open_zones[open_zone_idx]; |
1110 | if (new_zone_idx == zone_idx) | |
1111 | break; | |
1112 | zone_idx = new_zone_idx; | |
1113 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1114 | pthread_mutex_unlock(&z->mutex); | |
1115 | } | |
1116 | ||
1117 | /* Both z->mutex and f->zbd_info->mutex are held. */ | |
1118 | ||
1119 | examine_zone: | |
236d23a8 | 1120 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) { |
59b07544 BVA |
1121 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1122 | goto out; | |
1123 | } | |
b2da58c4 SK |
1124 | |
1125 | open_other_zone: | |
1126 | /* Check if number of open zones reaches one of limits. */ | |
1127 | wait_zone_close = | |
1128 | f->zbd_info->num_open_zones == f->max_zone - f->min_zone || | |
1129 | (td->o.max_open_zones && | |
1130 | f->zbd_info->num_open_zones == td->o.max_open_zones) || | |
1131 | (td->o.job_max_open_zones && | |
1132 | td->num_open_zones == td->o.job_max_open_zones); | |
1133 | ||
59b07544 BVA |
1134 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1135 | ||
1136 | /* Only z->mutex is held. */ | |
1137 | ||
b2da58c4 SK |
1138 | /* |
1139 | * When number of open zones reaches to one of limits, wait for | |
1140 | * zone close before opening a new zone. | |
1141 | */ | |
1142 | if (wait_zone_close) { | |
1143 | dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n", | |
1144 | __func__, f->file_name); | |
1145 | io_u_quiesce(td); | |
1146 | } | |
1147 | ||
59b07544 BVA |
1148 | /* Zone 'z' is full, so try to open a new zone. */ |
1149 | for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) { | |
1150 | zone_idx++; | |
1151 | pthread_mutex_unlock(&z->mutex); | |
1152 | z++; | |
ee3696bd | 1153 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1154 | /* Wrap-around. */ |
fae3b9a0 | 1155 | zone_idx = f->min_zone; |
923f7c1e | 1156 | z = get_zone(f, zone_idx); |
59b07544 | 1157 | } |
ee3696bd | 1158 | assert(is_valid_offset(f, z->start)); |
fae3b9a0 | 1159 | zone_lock(td, f, z); |
59b07544 BVA |
1160 | if (z->open) |
1161 | continue; | |
954217b9 | 1162 | if (zbd_open_zone(td, f, zone_idx)) |
59b07544 BVA |
1163 | goto out; |
1164 | } | |
1165 | ||
1166 | /* Only z->mutex is held. */ | |
1167 | ||
1168 | /* Check whether the write fits in any of the already opened zones. */ | |
1169 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1170 | for (i = 0; i < f->zbd_info->num_open_zones; i++) { | |
1171 | zone_idx = f->zbd_info->open_zones[i]; | |
fae3b9a0 AD |
1172 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1173 | continue; | |
59b07544 BVA |
1174 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1175 | pthread_mutex_unlock(&z->mutex); | |
1176 | ||
923f7c1e | 1177 | z = get_zone(f, zone_idx); |
59b07544 | 1178 | |
fae3b9a0 | 1179 | zone_lock(td, f, z); |
236d23a8 | 1180 | if (z->wp + min_bs <= zbd_zone_capacity_end(z)) |
59b07544 BVA |
1181 | goto out; |
1182 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1183 | } | |
1184 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1185 | pthread_mutex_unlock(&z->mutex); | |
1186 | dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__, | |
1187 | f->file_name); | |
1188 | return NULL; | |
1189 | ||
1190 | out: | |
1191 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name, | |
1192 | zone_idx); | |
ee3696bd | 1193 | io_u->offset = z->start; |
59b07544 BVA |
1194 | return z; |
1195 | } | |
1196 | ||
bfbdd35b | 1197 | /* The caller must hold z->mutex. */ |
59b07544 BVA |
1198 | static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, |
1199 | struct io_u *io_u, | |
1200 | struct fio_zone_info *z) | |
bfbdd35b BVA |
1201 | { |
1202 | const struct fio_file *f = io_u->file; | |
1203 | const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
1204 | ||
122c3539 | 1205 | if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) { |
59b07544 BVA |
1206 | pthread_mutex_unlock(&z->mutex); |
1207 | z = zbd_convert_to_open_zone(td, io_u); | |
1208 | assert(z); | |
1209 | } | |
1210 | ||
236d23a8 | 1211 | if (z->verify_block * min_bs >= z->capacity) |
a0c84dd4 | 1212 | log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block, |
236d23a8 | 1213 | min_bs, (unsigned long long)z->capacity); |
ee3696bd | 1214 | io_u->offset = z->start + z->verify_block++ * min_bs; |
59b07544 | 1215 | return z; |
bfbdd35b BVA |
1216 | } |
1217 | ||
1218 | /* | |
1219 | * Find another zone for which @io_u fits below the write pointer. Start | |
1220 | * searching in zones @zb + 1 .. @zl and continue searching in zones | |
1221 | * @zf .. @zb - 1. | |
1222 | * | |
1223 | * Either returns NULL or returns a zone pointer and holds the mutex for that | |
1224 | * zone. | |
1225 | */ | |
1226 | static struct fio_zone_info * | |
1227 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, | |
1228 | struct fio_zone_info *zb, struct fio_zone_info *zl) | |
1229 | { | |
1230 | const uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
fae3b9a0 | 1231 | struct fio_file *f = io_u->file; |
bfbdd35b | 1232 | struct fio_zone_info *z1, *z2; |
923f7c1e | 1233 | const struct fio_zone_info *const zf = get_zone(f, f->min_zone); |
bfbdd35b BVA |
1234 | |
1235 | /* | |
1236 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1237 | * the nearest non-empty zone in case of random I/O. | |
1238 | */ | |
1239 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1240 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
fae3b9a0 | 1241 | zone_lock(td, f, z1); |
ee3696bd | 1242 | if (z1->start + min_bs <= z1->wp) |
bfbdd35b BVA |
1243 | return z1; |
1244 | pthread_mutex_unlock(&z1->mutex); | |
1245 | } else if (!td_random(td)) { | |
1246 | break; | |
1247 | } | |
1248 | if (td_random(td) && z2 >= zf && | |
b7694961 | 1249 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
fae3b9a0 | 1250 | zone_lock(td, f, z2); |
ee3696bd | 1251 | if (z2->start + min_bs <= z2->wp) |
bfbdd35b BVA |
1252 | return z2; |
1253 | pthread_mutex_unlock(&z2->mutex); | |
1254 | } | |
1255 | } | |
1256 | dprint(FD_ZBD, "%s: adjusting random read offset failed\n", | |
1257 | f->file_name); | |
1258 | return NULL; | |
1259 | } | |
1260 | ||
b2da58c4 SK |
1261 | /** |
1262 | * zbd_end_zone_io - update zone status at command completion | |
1263 | * @io_u: I/O unit | |
1264 | * @z: zone info pointer | |
1265 | * | |
1266 | * If the write command made the zone full, close it. | |
1267 | * | |
1268 | * The caller must hold z->mutex. | |
1269 | */ | |
1270 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1271 | struct fio_zone_info *z) | |
1272 | { | |
1273 | const struct fio_file *f = io_u->file; | |
1274 | ||
1275 | if (io_u->ddir == DDIR_WRITE && | |
1276 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1277 | pthread_mutex_lock(&f->zbd_info->mutex); | |
122c3539 | 1278 | zbd_close_zone(td, f, zbd_zone_nr(f, z)); |
b2da58c4 SK |
1279 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1280 | } | |
1281 | } | |
1282 | ||
bfbdd35b | 1283 | /** |
d9ed3e63 | 1284 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1285 | * @io_u: I/O unit |
d9ed3e63 DLM |
1286 | * @success: Whether or not the I/O unit has been queued successfully |
1287 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1288 | * |
d9ed3e63 DLM |
1289 | * For write and trim operations, update the write pointer of the I/O unit |
1290 | * target zone. | |
bfbdd35b | 1291 | */ |
b2da58c4 SK |
1292 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1293 | bool success) | |
bfbdd35b | 1294 | { |
d9ed3e63 DLM |
1295 | const struct fio_file *f = io_u->file; |
1296 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b BVA |
1297 | struct fio_zone_info *z; |
1298 | uint32_t zone_idx; | |
d9ed3e63 | 1299 | uint64_t zone_end; |
bfbdd35b | 1300 | |
bfbdd35b BVA |
1301 | if (!zbd_info) |
1302 | return; | |
1303 | ||
d9ed3e63 | 1304 | zone_idx = zbd_zone_idx(f, io_u->offset); |
bfbdd35b | 1305 | assert(zone_idx < zbd_info->nr_zones); |
923f7c1e | 1306 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1307 | |
b7694961 | 1308 | if (!zbd_zone_swr(z)) |
bfbdd35b | 1309 | return; |
d9ed3e63 | 1310 | |
bfbdd35b BVA |
1311 | if (!success) |
1312 | goto unlock; | |
d9ed3e63 DLM |
1313 | |
1314 | dprint(FD_ZBD, | |
1315 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
1316 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1317 | ||
bfbdd35b BVA |
1318 | switch (io_u->ddir) { |
1319 | case DDIR_WRITE: | |
d9ed3e63 | 1320 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1321 | zbd_zone_capacity_end(z)); |
a7c2b6fc BVA |
1322 | pthread_mutex_lock(&zbd_info->mutex); |
1323 | /* | |
1324 | * z->wp > zone_end means that one or more I/O errors | |
1325 | * have occurred. | |
1326 | */ | |
1327 | if (z->wp <= zone_end) | |
1328 | zbd_info->sectors_with_data += zone_end - z->wp; | |
1329 | pthread_mutex_unlock(&zbd_info->mutex); | |
bfbdd35b BVA |
1330 | z->wp = zone_end; |
1331 | break; | |
1332 | case DDIR_TRIM: | |
1333 | assert(z->wp == z->start); | |
1334 | break; | |
1335 | default: | |
1336 | break; | |
1337 | } | |
d9ed3e63 | 1338 | |
b2da58c4 SK |
1339 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1340 | zbd_end_zone_io(td, io_u, z); | |
1341 | ||
bfbdd35b | 1342 | unlock: |
d9ed3e63 DLM |
1343 | if (!success || q != FIO_Q_QUEUED) { |
1344 | /* BUSY or COMPLETED: unlock the zone */ | |
1345 | pthread_mutex_unlock(&z->mutex); | |
1346 | io_u->zbd_put_io = NULL; | |
1347 | } | |
1348 | } | |
1349 | ||
1350 | /** | |
1351 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1352 | * @io_u: I/O unit | |
1353 | */ | |
b2da58c4 | 1354 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1355 | { |
1356 | const struct fio_file *f = io_u->file; | |
1357 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1358 | struct fio_zone_info *z; | |
1359 | uint32_t zone_idx; | |
f59c2f8e | 1360 | int ret; |
d9ed3e63 DLM |
1361 | |
1362 | if (!zbd_info) | |
1363 | return; | |
615555bb | 1364 | |
d9ed3e63 DLM |
1365 | zone_idx = zbd_zone_idx(f, io_u->offset); |
1366 | assert(zone_idx < zbd_info->nr_zones); | |
923f7c1e | 1367 | z = get_zone(f, zone_idx); |
d9ed3e63 | 1368 | |
b7694961 | 1369 | if (!zbd_zone_swr(z)) |
d9ed3e63 DLM |
1370 | return; |
1371 | ||
1372 | dprint(FD_ZBD, | |
1373 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
1374 | f->file_name, io_u->offset, io_u->buflen, zone_idx); | |
1375 | ||
b2da58c4 SK |
1376 | zbd_end_zone_io(td, io_u, z); |
1377 | ||
f59c2f8e AD |
1378 | ret = pthread_mutex_unlock(&z->mutex); |
1379 | assert(ret == 0); | |
d9ed3e63 | 1380 | zbd_check_swd(f); |
bfbdd35b BVA |
1381 | } |
1382 | ||
9d87c646 DLM |
1383 | /* |
1384 | * Windows and MacOS do not define this. | |
1385 | */ | |
1386 | #ifndef EREMOTEIO | |
1387 | #define EREMOTEIO 121 /* POSIX value */ | |
1388 | #endif | |
1389 | ||
bfbdd35b BVA |
1390 | bool zbd_unaligned_write(int error_code) |
1391 | { | |
1392 | switch (error_code) { | |
1393 | case EIO: | |
1394 | case EREMOTEIO: | |
1395 | return true; | |
1396 | } | |
1397 | return false; | |
1398 | } | |
1399 | ||
4d37720a DLM |
1400 | /** |
1401 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1402 | * @td: FIO thread data. | |
1403 | * @io_u: FIO I/O unit. | |
1404 | * | |
1405 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1406 | * no more IO can be performed in the current zone. | |
1407 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1408 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1409 | * - For write workloads, zoneskip is applied when the zone is full. | |
1410 | * This applies only to read and write operations. | |
1411 | */ | |
1412 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1413 | { | |
1414 | struct fio_file *f = io_u->file; | |
1415 | enum fio_ddir ddir = io_u->ddir; | |
1416 | struct fio_zone_info *z; | |
1417 | uint32_t zone_idx; | |
1418 | ||
1419 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1420 | assert(td->o.zone_size); | |
1421 | ||
236d23a8 | 1422 | zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); |
923f7c1e | 1423 | z = get_zone(f, zone_idx); |
236d23a8 SK |
1424 | |
1425 | /* | |
1426 | * When the zone capacity is smaller than the zone size and the I/O is | |
1427 | * sequential write, skip to zone end if the latest position is at the | |
1428 | * zone capacity limit. | |
1429 | */ | |
1430 | if (z->capacity < f->zbd_info->zone_size && !td_random(td) && | |
1431 | ddir == DDIR_WRITE && | |
1432 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { | |
1433 | dprint(FD_ZBD, | |
1434 | "%s: Jump from zone capacity limit to zone end:" | |
5090d1d0 JA |
1435 | " (%llu -> %llu) for zone %u (%llu)\n", |
1436 | f->file_name, (unsigned long long) f->last_pos[ddir], | |
122c3539 | 1437 | (unsigned long long) zbd_zone_end(z), zone_idx, |
5090d1d0 | 1438 | (unsigned long long) z->capacity); |
236d23a8 SK |
1439 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1440 | f->last_pos[ddir] = zbd_zone_end(z); | |
1441 | } | |
1442 | ||
4d37720a DLM |
1443 | /* |
1444 | * zone_skip is valid only for sequential workloads. | |
1445 | */ | |
1446 | if (td_random(td) || !td->o.zone_skip) | |
1447 | return; | |
1448 | ||
1449 | /* | |
1450 | * It is time to switch to a new zone if: | |
1451 | * - zone_bytes == zone_size bytes have already been accessed | |
1452 | * - The last position reached the end of the current zone. | |
1453 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1454 | * reached the zone write pointer. | |
1455 | */ | |
4d37720a | 1456 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1457 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1458 | (ddir == DDIR_READ && |
1459 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1460 | /* | |
1461 | * Skip zones. | |
1462 | */ | |
1463 | td->zone_bytes = 0; | |
1464 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1465 | ||
1466 | /* | |
1467 | * Wrap from the beginning, if we exceed the file size | |
1468 | */ | |
1469 | if (f->file_offset >= f->real_file_size) | |
1470 | f->file_offset = get_start_offset(td, f); | |
1471 | ||
1472 | f->last_pos[ddir] = f->file_offset; | |
1473 | td->io_skip_bytes += td->o.zone_skip; | |
1474 | } | |
1475 | } | |
1476 | ||
c65057f9 | 1477 | /** |
c7d5e152 | 1478 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1479 | * |
1480 | * @td: FIO thread data. | |
1481 | * @io_u: FIO I/O unit. | |
1482 | * @ddir: I/O direction before adjustment. | |
1483 | * | |
1484 | * Return adjusted I/O direction. | |
1485 | */ | |
1486 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1487 | enum fio_ddir ddir) | |
1488 | { | |
1489 | /* | |
1490 | * In case read direction is chosen for the first random I/O, fio with | |
1491 | * zonemode=zbd stops because no data can be read from zoned block | |
1492 | * devices with all empty zones. Overwrite the first I/O direction as | |
1493 | * write to make sure data to read exists. | |
1494 | */ | |
731461cc | 1495 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1496 | return ddir; |
1497 | ||
1498 | if (io_u->file->zbd_info->sectors_with_data || | |
1499 | td->o.read_beyond_wp) | |
1500 | return DDIR_READ; | |
1501 | ||
1502 | return DDIR_WRITE; | |
1503 | } | |
1504 | ||
bfbdd35b BVA |
1505 | /** |
1506 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1507 | * @td: FIO thread data. | |
1508 | * @io_u: FIO I/O unit. | |
1509 | * | |
1510 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1511 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1512 | * corresponds to io_u->offset at the end of this function. | |
1513 | */ | |
1514 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1515 | { | |
b7694961 | 1516 | struct fio_file *f = io_u->file; |
bfbdd35b | 1517 | uint32_t zone_idx_b; |
de65f7b7 | 1518 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b BVA |
1519 | uint32_t orig_len = io_u->buflen; |
1520 | uint32_t min_bs = td->o.min_bs[io_u->ddir]; | |
1521 | uint64_t new_len; | |
1522 | int64_t range; | |
1523 | ||
1524 | if (!f->zbd_info) | |
1525 | return io_u_accept; | |
1526 | ||
adc6adcb | 1527 | assert(min_bs); |
bfbdd35b BVA |
1528 | assert(is_valid_offset(f, io_u->offset)); |
1529 | assert(io_u->buflen); | |
1530 | zone_idx_b = zbd_zone_idx(f, io_u->offset); | |
923f7c1e | 1531 | zb = get_zone(f, zone_idx_b); |
de65f7b7 | 1532 | orig_zb = zb; |
bfbdd35b BVA |
1533 | |
1534 | /* Accept the I/O offset for conventional zones. */ | |
b7694961 | 1535 | if (!zbd_zone_swr(zb)) |
bfbdd35b BVA |
1536 | return io_u_accept; |
1537 | ||
1538 | /* | |
1539 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1540 | * is enabled. | |
1541 | */ | |
b7694961 | 1542 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1543 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1544 | return io_u_accept; | |
1545 | ||
615555bb BVA |
1546 | zbd_check_swd(f); |
1547 | ||
fae3b9a0 | 1548 | zone_lock(td, f, zb); |
6f0c6085 | 1549 | |
bfbdd35b BVA |
1550 | switch (io_u->ddir) { |
1551 | case DDIR_READ: | |
0ed7d55e AR |
1552 | if (td->runstate == TD_VERIFYING && td_write(td)) { |
1553 | zb = zbd_replay_write_order(td, io_u, zb); | |
b2726d53 | 1554 | pthread_mutex_unlock(&zb->mutex); |
bfbdd35b BVA |
1555 | goto accept; |
1556 | } | |
1557 | /* | |
de65f7b7 DLM |
1558 | * Check that there is enough written data in the zone to do an |
1559 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1560 | * the I/O. | |
bfbdd35b | 1561 | */ |
b7694961 | 1562 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1563 | zb->wp - zb->start : 0; |
de65f7b7 | 1564 | if (range < min_bs || |
ee3696bd | 1565 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
bfbdd35b | 1566 | pthread_mutex_unlock(&zb->mutex); |
923f7c1e | 1567 | zl = get_zone(f, f->max_zone); |
bfbdd35b BVA |
1568 | zb = zbd_find_zone(td, io_u, zb, zl); |
1569 | if (!zb) { | |
1570 | dprint(FD_ZBD, | |
1571 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1572 | f->file_name, io_u->offset, | |
1573 | io_u->buflen); | |
1574 | goto eof; | |
1575 | } | |
de65f7b7 DLM |
1576 | /* |
1577 | * zbd_find_zone() returned a zone with a range of at | |
1578 | * least min_bs. | |
1579 | */ | |
ee3696bd | 1580 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1581 | assert(range >= min_bs); |
1582 | ||
1583 | if (!td_random(td)) | |
ee3696bd | 1584 | io_u->offset = zb->start; |
bfbdd35b | 1585 | } |
de65f7b7 DLM |
1586 | /* |
1587 | * Make sure the I/O is within the zone valid data range while | |
1588 | * maximizing the I/O size and preserving randomness. | |
1589 | */ | |
1590 | if (range <= io_u->buflen) | |
ee3696bd | 1591 | io_u->offset = zb->start; |
de65f7b7 | 1592 | else if (td_random(td)) |
ee3696bd DLM |
1593 | io_u->offset = zb->start + |
1594 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 DLM |
1595 | (range - io_u->buflen)) / min_bs * min_bs; |
1596 | /* | |
1597 | * Make sure the I/O does not cross over the zone wp position. | |
1598 | */ | |
1599 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1600 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1601 | new_len = new_len / min_bs * min_bs; |
1602 | if (new_len < io_u->buflen) { | |
1603 | io_u->buflen = new_len; | |
1604 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1605 | orig_len, io_u->buflen); | |
bfbdd35b | 1606 | } |
ee3696bd DLM |
1607 | assert(zb->start <= io_u->offset); |
1608 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
bfbdd35b BVA |
1609 | goto accept; |
1610 | case DDIR_WRITE: | |
ee3696bd | 1611 | if (io_u->buflen > f->zbd_info->zone_size) |
bfbdd35b | 1612 | goto eof; |
954217b9 | 1613 | if (!zbd_open_zone(td, f, zone_idx_b)) { |
59b07544 BVA |
1614 | pthread_mutex_unlock(&zb->mutex); |
1615 | zb = zbd_convert_to_open_zone(td, io_u); | |
1616 | if (!zb) | |
1617 | goto eof; | |
122c3539 | 1618 | zone_idx_b = zbd_zone_nr(f, zb); |
59b07544 | 1619 | } |
a7c2b6fc BVA |
1620 | /* Check whether the zone reset threshold has been exceeded */ |
1621 | if (td->o.zrf.u.f) { | |
ee3696bd | 1622 | if (f->zbd_info->sectors_with_data >= |
a7c2b6fc BVA |
1623 | f->io_size * td->o.zrt.u.f && |
1624 | zbd_dec_and_reset_write_cnt(td, f)) { | |
1625 | zb->reset_zone = 1; | |
1626 | } | |
1627 | } | |
bfbdd35b BVA |
1628 | /* Reset the zone pointer if necessary */ |
1629 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
1630 | assert(td->o.verify == VERIFY_NONE); | |
1631 | /* | |
1632 | * Since previous write requests may have been submitted | |
1633 | * asynchronously and since we will submit the zone | |
1634 | * reset synchronously, wait until previously submitted | |
1635 | * write requests have completed before issuing a | |
1636 | * zone reset. | |
1637 | */ | |
1638 | io_u_quiesce(td); | |
1639 | zb->reset_zone = 0; | |
1640 | if (zbd_reset_zone(td, f, zb) < 0) | |
1641 | goto eof; | |
236d23a8 SK |
1642 | |
1643 | if (zb->capacity < min_bs) { | |
1644 | log_err("zone capacity %llu smaller than minimum block size %d\n", | |
1645 | (unsigned long long)zb->capacity, | |
1646 | min_bs); | |
1647 | goto eof; | |
1648 | } | |
bfbdd35b BVA |
1649 | } |
1650 | /* Make writes occur at the write pointer */ | |
1651 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 1652 | io_u->offset = zb->wp; |
bfbdd35b BVA |
1653 | if (!is_valid_offset(f, io_u->offset)) { |
1654 | dprint(FD_ZBD, "Dropped request with offset %llu\n", | |
1655 | io_u->offset); | |
1656 | goto eof; | |
1657 | } | |
1658 | /* | |
1659 | * Make sure that the buflen is a multiple of the minimal | |
1660 | * block size. Give up if shrinking would make the request too | |
1661 | * small. | |
1662 | */ | |
1663 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 1664 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
1665 | new_len = new_len / min_bs * min_bs; |
1666 | if (new_len == io_u->buflen) | |
1667 | goto accept; | |
1668 | if (new_len >= min_bs) { | |
1669 | io_u->buflen = new_len; | |
1670 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1671 | orig_len, io_u->buflen); | |
1672 | goto accept; | |
1673 | } | |
1674 | log_err("Zone remainder %lld smaller than minimum block size %d\n", | |
236d23a8 | 1675 | (zbd_zone_capacity_end(zb) - io_u->offset), |
bfbdd35b BVA |
1676 | min_bs); |
1677 | goto eof; | |
1678 | case DDIR_TRIM: | |
1679 | /* fall-through */ | |
1680 | case DDIR_SYNC: | |
1681 | case DDIR_DATASYNC: | |
1682 | case DDIR_SYNC_FILE_RANGE: | |
1683 | case DDIR_WAIT: | |
1684 | case DDIR_LAST: | |
1685 | case DDIR_INVAL: | |
1686 | goto accept; | |
1687 | } | |
1688 | ||
1689 | assert(false); | |
1690 | ||
1691 | accept: | |
1692 | assert(zb); | |
b7694961 | 1693 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
1694 | assert(!io_u->zbd_queue_io); |
1695 | assert(!io_u->zbd_put_io); | |
1696 | io_u->zbd_queue_io = zbd_queue_io; | |
1697 | io_u->zbd_put_io = zbd_put_io; | |
bfbdd35b BVA |
1698 | return io_u_accept; |
1699 | ||
1700 | eof: | |
1701 | if (zb) | |
1702 | pthread_mutex_unlock(&zb->mutex); | |
1703 | return io_u_eof; | |
1704 | } | |
fd5d733f BVA |
1705 | |
1706 | /* Return a string with ZBD statistics */ | |
1707 | char *zbd_write_status(const struct thread_stat *ts) | |
1708 | { | |
1709 | char *res; | |
1710 | ||
a0c84dd4 | 1711 | if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0) |
fd5d733f BVA |
1712 | return NULL; |
1713 | return res; | |
1714 | } |