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