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 | ||
410a071c DLM |
25 | static bool is_valid_offset(const struct fio_file *f, uint64_t offset) |
26 | { | |
27 | return (uint64_t)(offset - f->file_offset) < f->io_size; | |
28 | } | |
29 | ||
dc8a3d62 DLM |
30 | static inline unsigned int zbd_zone_idx(const struct fio_file *f, |
31 | struct fio_zone_info *zone) | |
410a071c DLM |
32 | { |
33 | return zone - f->zbd_info->zone_info; | |
34 | } | |
35 | ||
36 | /** | |
dc8a3d62 | 37 | * zbd_offset_to_zone_idx - convert an offset into a zone number |
410a071c DLM |
38 | * @f: file pointer. |
39 | * @offset: offset in bytes. If this offset is in the first zone_size bytes | |
40 | * past the disk size then the index of the sentinel is returned. | |
41 | */ | |
dc8a3d62 DLM |
42 | static unsigned int zbd_offset_to_zone_idx(const struct fio_file *f, |
43 | uint64_t offset) | |
410a071c DLM |
44 | { |
45 | uint32_t zone_idx; | |
46 | ||
47 | if (f->zbd_info->zone_size_log2 > 0) | |
48 | zone_idx = offset >> f->zbd_info->zone_size_log2; | |
49 | else | |
50 | zone_idx = offset / f->zbd_info->zone_size; | |
51 | ||
52 | return min(zone_idx, f->zbd_info->nr_zones); | |
53 | } | |
54 | ||
55 | /** | |
56 | * zbd_zone_end - Return zone end location | |
57 | * @z: zone info pointer. | |
58 | */ | |
59 | static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) | |
60 | { | |
61 | return (z+1)->start; | |
62 | } | |
63 | ||
64 | /** | |
65 | * zbd_zone_capacity_end - Return zone capacity limit end location | |
66 | * @z: zone info pointer. | |
67 | */ | |
68 | static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) | |
69 | { | |
70 | return z->start + z->capacity; | |
71 | } | |
72 | ||
df67bf1e SK |
73 | /** |
74 | * zbd_zone_remainder - Return the number of bytes that are still available for | |
75 | * writing before the zone gets full | |
76 | * @z: zone info pointer. | |
77 | */ | |
78 | static inline uint64_t zbd_zone_remainder(struct fio_zone_info *z) | |
79 | { | |
80 | if (z->wp >= zbd_zone_capacity_end(z)) | |
81 | return 0; | |
82 | ||
83 | return zbd_zone_capacity_end(z) - z->wp; | |
84 | } | |
85 | ||
410a071c DLM |
86 | /** |
87 | * zbd_zone_full - verify whether a minimum number of bytes remain in a zone | |
88 | * @f: file pointer. | |
89 | * @z: zone info pointer. | |
90 | * @required: minimum number of bytes that must remain in a zone. | |
91 | * | |
92 | * The caller must hold z->mutex. | |
93 | */ | |
94 | static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, | |
95 | uint64_t required) | |
96 | { | |
97 | assert((required & 511) == 0); | |
98 | ||
df67bf1e | 99 | return z->has_wp && required > zbd_zone_remainder(z); |
410a071c DLM |
100 | } |
101 | ||
102 | static void zone_lock(struct thread_data *td, const struct fio_file *f, | |
103 | struct fio_zone_info *z) | |
104 | { | |
105 | struct zoned_block_device_info *zbd = f->zbd_info; | |
106 | uint32_t nz = z - zbd->zone_info; | |
107 | ||
108 | /* A thread should never lock zones outside its working area. */ | |
109 | assert(f->min_zone <= nz && nz < f->max_zone); | |
110 | ||
111 | assert(z->has_wp); | |
112 | ||
113 | /* | |
114 | * Lock the io_u target zone. The zone will be unlocked if io_u offset | |
115 | * is changed or when io_u completes and zbd_put_io() executed. | |
116 | * To avoid multiple jobs doing asynchronous I/Os from deadlocking each | |
117 | * other waiting for zone locks when building an io_u batch, first | |
118 | * only trylock the zone. If the zone is already locked by another job, | |
119 | * process the currently queued I/Os so that I/O progress is made and | |
120 | * zones unlocked. | |
121 | */ | |
122 | if (pthread_mutex_trylock(&z->mutex) != 0) { | |
123 | if (!td_ioengine_flagged(td, FIO_SYNCIO)) | |
124 | io_u_quiesce(td); | |
125 | pthread_mutex_lock(&z->mutex); | |
126 | } | |
127 | } | |
128 | ||
129 | static inline void zone_unlock(struct fio_zone_info *z) | |
130 | { | |
131 | int ret; | |
132 | ||
133 | assert(z->has_wp); | |
134 | ret = pthread_mutex_unlock(&z->mutex); | |
135 | assert(!ret); | |
136 | } | |
137 | ||
39e06ee7 DLM |
138 | static inline struct fio_zone_info *zbd_get_zone(const struct fio_file *f, |
139 | unsigned int zone_idx) | |
410a071c | 140 | { |
39e06ee7 | 141 | return &f->zbd_info->zone_info[zone_idx]; |
410a071c DLM |
142 | } |
143 | ||
53aa6171 DLM |
144 | static inline struct fio_zone_info * |
145 | zbd_offset_to_zone(const struct fio_file *f, uint64_t offset) | |
146 | { | |
147 | return zbd_get_zone(f, zbd_offset_to_zone_idx(f, offset)); | |
148 | } | |
149 | ||
2fb29f27 SK |
150 | static bool accounting_vdb(struct thread_data *td, const struct fio_file *f) |
151 | { | |
152 | return td->o.zrt.u.f && td_write(td); | |
153 | } | |
154 | ||
b7694961 DLM |
155 | /** |
156 | * zbd_get_zoned_model - Get a device zoned model | |
157 | * @td: FIO thread data | |
158 | * @f: FIO file for which to get model information | |
159 | */ | |
38334c13 DLM |
160 | static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, |
161 | enum zbd_zoned_model *model) | |
b7694961 DLM |
162 | { |
163 | int ret; | |
164 | ||
50cc48d5 NC |
165 | if (f->filetype == FIO_TYPE_PIPE) { |
166 | log_err("zonemode=zbd does not support pipes\n"); | |
167 | return -EINVAL; | |
168 | } | |
169 | ||
9db0cde8 NC |
170 | /* If regular file, always emulate zones inside the file. */ |
171 | if (f->filetype == FIO_TYPE_FILE) { | |
172 | *model = ZBD_NONE; | |
173 | return 0; | |
174 | } | |
175 | ||
6c5b11d3 DLM |
176 | if (td->io_ops && td->io_ops->get_zoned_model) |
177 | ret = td->io_ops->get_zoned_model(td, f, model); | |
178 | else | |
179 | ret = blkzoned_get_zoned_model(td, f, model); | |
b7694961 DLM |
180 | if (ret < 0) { |
181 | td_verror(td, errno, "get zoned model failed"); | |
182 | log_err("%s: get zoned model failed (%d).\n", | |
183 | f->file_name, errno); | |
184 | } | |
185 | ||
186 | return ret; | |
187 | } | |
188 | ||
189 | /** | |
190 | * zbd_report_zones - Get zone information | |
191 | * @td: FIO thread data. | |
192 | * @f: FIO file for which to get zone information | |
193 | * @offset: offset from which to report zones | |
194 | * @zones: Array of struct zbd_zone | |
195 | * @nr_zones: Size of @zones array | |
196 | * | |
197 | * Get zone information into @zones starting from the zone at offset @offset | |
198 | * for the device specified by @f. | |
199 | * | |
200 | * Returns the number of zones reported upon success and a negative error code | |
201 | * upon failure. If the zone report is empty, always assume an error (device | |
202 | * problem) and return -EIO. | |
203 | */ | |
38334c13 DLM |
204 | static int zbd_report_zones(struct thread_data *td, struct fio_file *f, |
205 | uint64_t offset, struct zbd_zone *zones, | |
206 | unsigned int nr_zones) | |
b7694961 DLM |
207 | { |
208 | int ret; | |
209 | ||
6c5b11d3 DLM |
210 | if (td->io_ops && td->io_ops->report_zones) |
211 | ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); | |
212 | else | |
213 | ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); | |
b7694961 DLM |
214 | if (ret < 0) { |
215 | td_verror(td, errno, "report zones failed"); | |
362ce037 BVA |
216 | log_err("%s: report zones from sector %"PRIu64" failed (nr_zones=%d; errno=%d).\n", |
217 | f->file_name, offset >> 9, nr_zones, errno); | |
b7694961 DLM |
218 | } else if (ret == 0) { |
219 | td_verror(td, errno, "Empty zone report"); | |
ee5e3436 SK |
220 | log_err("%s: report zones from sector %"PRIu64" is empty.\n", |
221 | f->file_name, offset >> 9); | |
b7694961 DLM |
222 | ret = -EIO; |
223 | } | |
224 | ||
225 | return ret; | |
226 | } | |
227 | ||
228 | /** | |
229 | * zbd_reset_wp - reset the write pointer of a range of zones | |
230 | * @td: FIO thread data. | |
231 | * @f: FIO file for which to reset zones | |
232 | * @offset: Starting offset of the first zone to reset | |
233 | * @length: Length of the range of zones to reset | |
234 | * | |
235 | * Reset the write pointer of all zones in the range @offset...@offset+@length. | |
236 | * Returns 0 upon success and a negative error code upon failure. | |
237 | */ | |
38334c13 DLM |
238 | static int zbd_reset_wp(struct thread_data *td, struct fio_file *f, |
239 | uint64_t offset, uint64_t length) | |
b7694961 DLM |
240 | { |
241 | int ret; | |
242 | ||
6c5b11d3 DLM |
243 | if (td->io_ops && td->io_ops->reset_wp) |
244 | ret = td->io_ops->reset_wp(td, f, offset, length); | |
245 | else | |
246 | ret = blkzoned_reset_wp(td, f, offset, length); | |
b7694961 DLM |
247 | if (ret < 0) { |
248 | td_verror(td, errno, "resetting wp failed"); | |
ee5e3436 SK |
249 | log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n", |
250 | f->file_name, length >> 9, offset >> 9, errno); | |
b7694961 DLM |
251 | } |
252 | ||
253 | return ret; | |
254 | } | |
255 | ||
410a071c DLM |
256 | /** |
257 | * zbd_reset_zone - reset the write pointer of a single zone | |
258 | * @td: FIO thread data. | |
259 | * @f: FIO file associated with the disk for which to reset a write pointer. | |
260 | * @z: Zone to reset. | |
261 | * | |
262 | * Returns 0 upon success and a negative error code upon failure. | |
263 | * | |
264 | * The caller must hold z->mutex. | |
265 | */ | |
266 | static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, | |
267 | struct fio_zone_info *z) | |
268 | { | |
269 | uint64_t offset = z->start; | |
270 | uint64_t length = (z+1)->start - offset; | |
271 | uint64_t data_in_zone = z->wp - z->start; | |
272 | int ret = 0; | |
273 | ||
274 | if (!data_in_zone) | |
275 | return 0; | |
276 | ||
277 | assert(is_valid_offset(f, offset + length - 1)); | |
278 | ||
139d8dc6 | 279 | dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", |
dc8a3d62 | 280 | f->file_name, zbd_zone_idx(f, z)); |
139d8dc6 | 281 | |
410a071c DLM |
282 | switch (f->zbd_info->model) { |
283 | case ZBD_HOST_AWARE: | |
284 | case ZBD_HOST_MANAGED: | |
285 | ret = zbd_reset_wp(td, f, offset, length); | |
286 | if (ret < 0) | |
287 | return ret; | |
288 | break; | |
289 | default: | |
290 | break; | |
291 | } | |
292 | ||
2fb29f27 SK |
293 | if (accounting_vdb(td, f)) { |
294 | pthread_mutex_lock(&f->zbd_info->mutex); | |
295 | f->zbd_info->wp_valid_data_bytes -= data_in_zone; | |
296 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
297 | } | |
139d8dc6 | 298 | |
410a071c | 299 | z->wp = z->start; |
410a071c DLM |
300 | |
301 | td->ts.nr_zone_resets++; | |
302 | ||
303 | return ret; | |
304 | } | |
305 | ||
306 | /** | |
a4807046 | 307 | * zbd_write_zone_put - Remove a zone from the write target zones array. |
410a071c | 308 | * @td: FIO thread data. |
a4807046 | 309 | * @f: FIO file that has the write zones array to remove. |
410a071c DLM |
310 | * @zone_idx: Index of the zone to remove. |
311 | * | |
312 | * The caller must hold f->zbd_info->mutex. | |
313 | */ | |
a4807046 SK |
314 | static void zbd_write_zone_put(struct thread_data *td, const struct fio_file *f, |
315 | struct fio_zone_info *z) | |
410a071c | 316 | { |
a4807046 | 317 | uint32_t zi; |
410a071c | 318 | |
a4807046 | 319 | if (!z->write) |
a23411bb DLM |
320 | return; |
321 | ||
a4807046 SK |
322 | for (zi = 0; zi < f->zbd_info->num_write_zones; zi++) { |
323 | if (zbd_get_zone(f, f->zbd_info->write_zones[zi]) == z) | |
410a071c DLM |
324 | break; |
325 | } | |
a4807046 | 326 | if (zi == f->zbd_info->num_write_zones) |
410a071c DLM |
327 | return; |
328 | ||
a4807046 | 329 | dprint(FD_ZBD, "%s: removing zone %u from write zone array\n", |
dc8a3d62 | 330 | f->file_name, zbd_zone_idx(f, z)); |
139d8dc6 | 331 | |
a4807046 SK |
332 | memmove(f->zbd_info->write_zones + zi, |
333 | f->zbd_info->write_zones + zi + 1, | |
334 | (ZBD_MAX_WRITE_ZONES - (zi + 1)) * | |
335 | sizeof(f->zbd_info->write_zones[0])); | |
139d8dc6 | 336 | |
a4807046 SK |
337 | f->zbd_info->num_write_zones--; |
338 | td->num_write_zones--; | |
339 | z->write = 0; | |
410a071c DLM |
340 | } |
341 | ||
e1a1b59b SK |
342 | /** |
343 | * zbd_finish_zone - finish the specified zone | |
344 | * @td: FIO thread data. | |
345 | * @f: FIO file for which to finish a zone | |
346 | * @z: Zone to finish. | |
347 | * | |
348 | * Finish the zone at @offset with open or close status. | |
349 | */ | |
350 | static int zbd_finish_zone(struct thread_data *td, struct fio_file *f, | |
351 | struct fio_zone_info *z) | |
352 | { | |
353 | uint64_t offset = z->start; | |
354 | uint64_t length = f->zbd_info->zone_size; | |
355 | int ret = 0; | |
356 | ||
357 | switch (f->zbd_info->model) { | |
358 | case ZBD_HOST_AWARE: | |
359 | case ZBD_HOST_MANAGED: | |
360 | if (td->io_ops && td->io_ops->finish_zone) | |
361 | ret = td->io_ops->finish_zone(td, f, offset, length); | |
362 | else | |
363 | ret = blkzoned_finish_zone(td, f, offset, length); | |
364 | break; | |
365 | default: | |
366 | break; | |
367 | } | |
368 | ||
369 | if (ret < 0) { | |
370 | td_verror(td, errno, "finish zone failed"); | |
371 | log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n", | |
372 | f->file_name, offset >> 9, errno); | |
373 | } else { | |
374 | z->wp = (z+1)->start; | |
375 | } | |
376 | ||
377 | return ret; | |
378 | } | |
379 | ||
410a071c DLM |
380 | /** |
381 | * zbd_reset_zones - Reset a range of zones. | |
382 | * @td: fio thread data. | |
383 | * @f: fio file for which to reset zones | |
384 | * @zb: first zone to reset. | |
385 | * @ze: first zone not to reset. | |
386 | * | |
387 | * Returns 0 upon success and 1 upon failure. | |
388 | */ | |
389 | static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, | |
390 | struct fio_zone_info *const zb, | |
391 | struct fio_zone_info *const ze) | |
392 | { | |
393 | struct fio_zone_info *z; | |
394 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
395 | int res = 0; | |
396 | ||
397 | assert(min_bs); | |
398 | ||
139d8dc6 | 399 | dprint(FD_ZBD, "%s: examining zones %u .. %u\n", |
dc8a3d62 | 400 | f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze)); |
139d8dc6 | 401 | |
410a071c | 402 | for (z = zb; z < ze; z++) { |
410a071c DLM |
403 | if (!z->has_wp) |
404 | continue; | |
139d8dc6 | 405 | |
410a071c DLM |
406 | zone_lock(td, f, z); |
407 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a4807046 | 408 | zbd_write_zone_put(td, f, z); |
410a071c | 409 | pthread_mutex_unlock(&f->zbd_info->mutex); |
139d8dc6 | 410 | |
410a071c DLM |
411 | if (z->wp != z->start) { |
412 | dprint(FD_ZBD, "%s: resetting zone %u\n", | |
dc8a3d62 | 413 | f->file_name, zbd_zone_idx(f, z)); |
410a071c DLM |
414 | if (zbd_reset_zone(td, f, z) < 0) |
415 | res = 1; | |
416 | } | |
139d8dc6 | 417 | |
410a071c DLM |
418 | zone_unlock(z); |
419 | } | |
420 | ||
421 | return res; | |
422 | } | |
423 | ||
d2f442bc NC |
424 | /** |
425 | * zbd_get_max_open_zones - Get the maximum number of open zones | |
426 | * @td: FIO thread data | |
427 | * @f: FIO file for which to get max open zones | |
428 | * @max_open_zones: Upon success, result will be stored here. | |
429 | * | |
430 | * A @max_open_zones value set to zero means no limit. | |
431 | * | |
432 | * Returns 0 upon success and a negative error code upon failure. | |
433 | */ | |
38334c13 DLM |
434 | static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, |
435 | unsigned int *max_open_zones) | |
d2f442bc NC |
436 | { |
437 | int ret; | |
438 | ||
439 | if (td->io_ops && td->io_ops->get_max_open_zones) | |
440 | ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); | |
441 | else | |
442 | ret = blkzoned_get_max_open_zones(td, f, max_open_zones); | |
443 | if (ret < 0) { | |
444 | td_verror(td, errno, "get max open zones failed"); | |
445 | log_err("%s: get max open zones failed (%d).\n", | |
446 | f->file_name, errno); | |
447 | } | |
448 | ||
449 | return ret; | |
450 | } | |
451 | ||
bfbdd35b | 452 | /** |
f539b98c | 453 | * __zbd_write_zone_get - Add a zone to the array of write zones. |
410a071c | 454 | * @td: fio thread data. |
a4807046 | 455 | * @f: fio file that has the write zones array to add. |
410a071c | 456 | * @zone_idx: Index of the zone to add. |
bfbdd35b | 457 | * |
f539b98c SK |
458 | * Do same operation as @zbd_write_zone_get, except it adds the zone at |
459 | * @zone_idx to write target zones array even when it does not have remainder | |
460 | * space to write one block. | |
bfbdd35b | 461 | */ |
f539b98c SK |
462 | static bool __zbd_write_zone_get(struct thread_data *td, |
463 | const struct fio_file *f, | |
464 | struct fio_zone_info *z) | |
1f57803b | 465 | { |
410a071c | 466 | struct zoned_block_device_info *zbdi = f->zbd_info; |
dc8a3d62 | 467 | uint32_t zone_idx = zbd_zone_idx(f, z); |
410a071c | 468 | bool res = true; |
fae3b9a0 | 469 | |
410a071c DLM |
470 | if (z->cond == ZBD_ZONE_COND_OFFLINE) |
471 | return false; | |
43bcbd5b | 472 | |
1f57803b | 473 | /* |
410a071c DLM |
474 | * Skip full zones with data verification enabled because resetting a |
475 | * zone causes data loss and hence causes verification to fail. | |
1f57803b | 476 | */ |
f539b98c | 477 | if (td->o.verify != VERIFY_NONE && zbd_zone_remainder(z) == 0) |
410a071c | 478 | return false; |
4d4c71e6 | 479 | |
410a071c | 480 | /* |
a4807046 SK |
481 | * zbdi->max_write_zones == 0 means that there is no limit on the |
482 | * maximum number of write target zones. In this case, do no track write | |
483 | * target zones in zbdi->write_zones array. | |
410a071c | 484 | */ |
a4807046 | 485 | if (!zbdi->max_write_zones) |
410a071c | 486 | return true; |
4d4c71e6 | 487 | |
410a071c | 488 | pthread_mutex_lock(&zbdi->mutex); |
b5a0f7ce | 489 | |
a4807046 | 490 | if (z->write) { |
410a071c | 491 | /* |
b5a0f7ce | 492 | * If the zone is going to be completely filled by writes |
a4807046 SK |
493 | * already in-flight, handle it as a full zone instead of a |
494 | * write target zone. | |
410a071c | 495 | */ |
df67bf1e | 496 | if (!zbd_zone_remainder(z)) |
410a071c DLM |
497 | res = false; |
498 | goto out; | |
499 | } | |
139d8dc6 | 500 | |
410a071c DLM |
501 | res = false; |
502 | /* Zero means no limit */ | |
503 | if (td->o.job_max_open_zones > 0 && | |
a4807046 | 504 | td->num_write_zones >= td->o.job_max_open_zones) |
410a071c | 505 | goto out; |
a4807046 | 506 | if (zbdi->num_write_zones >= zbdi->max_write_zones) |
410a071c | 507 | goto out; |
139d8dc6 | 508 | |
a4807046 | 509 | dprint(FD_ZBD, "%s: adding zone %u to write zone array\n", |
139d8dc6 DLM |
510 | f->file_name, zone_idx); |
511 | ||
a4807046 SK |
512 | zbdi->write_zones[zbdi->num_write_zones++] = zone_idx; |
513 | td->num_write_zones++; | |
514 | z->write = 1; | |
410a071c | 515 | res = true; |
bfbdd35b | 516 | |
410a071c DLM |
517 | out: |
518 | pthread_mutex_unlock(&zbdi->mutex); | |
519 | return res; | |
923f7c1e DF |
520 | } |
521 | ||
f539b98c SK |
522 | /** |
523 | * zbd_write_zone_get - Add a zone to the array of write zones. | |
524 | * @td: fio thread data. | |
525 | * @f: fio file that has the open zones to add. | |
526 | * @zone_idx: Index of the zone to add. | |
527 | * | |
528 | * Add a ZBD zone to write target zones array, if it is not yet added. Returns | |
529 | * true if either the zone was already added or if the zone was successfully | |
530 | * added to the array without exceeding the maximum number of write zones. | |
531 | * Returns false if the zone was not already added and addition of the zone | |
532 | * would cause the zone limit to be exceeded. | |
533 | */ | |
534 | static bool zbd_write_zone_get(struct thread_data *td, const struct fio_file *f, | |
535 | struct fio_zone_info *z) | |
536 | { | |
537 | const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; | |
538 | ||
539 | /* | |
540 | * Skip full zones with data verification enabled because resetting a | |
541 | * zone causes data loss and hence causes verification to fail. | |
542 | */ | |
543 | if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) | |
544 | return false; | |
545 | ||
546 | return __zbd_write_zone_get(td, f, z); | |
547 | } | |
548 | ||
59c3200d | 549 | /* Verify whether direct I/O is used for all host-managed zoned block drives. */ |
bfbdd35b BVA |
550 | static bool zbd_using_direct_io(void) |
551 | { | |
bfbdd35b | 552 | struct fio_file *f; |
da8f124f | 553 | int j; |
bfbdd35b | 554 | |
da8f124f | 555 | for_each_td(td) { |
bfbdd35b BVA |
556 | if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) |
557 | continue; | |
558 | for_each_file(td, f, j) { | |
59c3200d | 559 | if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK && |
b7694961 | 560 | f->zbd_info->model == ZBD_HOST_MANAGED) |
bfbdd35b BVA |
561 | return false; |
562 | } | |
da8f124f | 563 | } end_for_each(); |
bfbdd35b BVA |
564 | |
565 | return true; | |
566 | } | |
567 | ||
568 | /* Whether or not the I/O range for f includes one or more sequential zones */ | |
b3e9bd03 | 569 | static bool zbd_is_seq_job(const struct fio_file *f) |
bfbdd35b BVA |
570 | { |
571 | uint32_t zone_idx, zone_idx_b, zone_idx_e; | |
572 | ||
573 | assert(f->zbd_info); | |
139d8dc6 | 574 | |
bfbdd35b BVA |
575 | if (f->io_size == 0) |
576 | return false; | |
139d8dc6 | 577 | |
dc8a3d62 DLM |
578 | zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset); |
579 | zone_idx_e = | |
580 | zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1); | |
bfbdd35b | 581 | for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) |
39e06ee7 | 582 | if (zbd_get_zone(f, zone_idx)->has_wp) |
bfbdd35b BVA |
583 | return true; |
584 | ||
585 | return false; | |
586 | } | |
587 | ||
0bf93a1a DLM |
588 | /* |
589 | * Verify whether the file offset and size parameters are aligned with zone | |
590 | * boundaries. If the file offset is not aligned, align it down to the start of | |
591 | * the zone containing the start offset and align up the file io_size parameter. | |
592 | */ | |
593 | static bool zbd_zone_align_file_sizes(struct thread_data *td, | |
594 | struct fio_file *f) | |
595 | { | |
596 | const struct fio_zone_info *z; | |
597 | uint64_t new_offset, new_end; | |
0bf93a1a DLM |
598 | |
599 | if (!f->zbd_info) | |
600 | return true; | |
601 | if (f->file_offset >= f->real_file_size) | |
602 | return true; | |
603 | if (!zbd_is_seq_job(f)) | |
604 | return true; | |
605 | ||
606 | if (!td->o.zone_size) { | |
607 | td->o.zone_size = f->zbd_info->zone_size; | |
608 | if (!td->o.zone_size) { | |
609 | log_err("%s: invalid 0 zone size\n", | |
610 | f->file_name); | |
611 | return false; | |
612 | } | |
613 | } else if (td->o.zone_size != f->zbd_info->zone_size) { | |
614 | log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n", | |
615 | f->file_name, td->o.zone_size, | |
616 | f->zbd_info->zone_size); | |
617 | return false; | |
618 | } | |
619 | ||
620 | if (td->o.zone_skip % td->o.zone_size) { | |
621 | log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", | |
622 | f->file_name, td->o.zone_skip, | |
623 | td->o.zone_size); | |
624 | return false; | |
625 | } | |
626 | ||
53aa6171 | 627 | z = zbd_offset_to_zone(f, f->file_offset); |
0bf93a1a DLM |
628 | if ((f->file_offset != z->start) && |
629 | (td->o.td_ddir != TD_DDIR_READ)) { | |
630 | new_offset = zbd_zone_end(z); | |
631 | if (new_offset >= f->file_offset + f->io_size) { | |
632 | log_info("%s: io_size must be at least one zone\n", | |
633 | f->file_name); | |
634 | return false; | |
635 | } | |
636 | log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n", | |
637 | f->file_name, f->file_offset, | |
638 | new_offset); | |
639 | f->io_size -= (new_offset - f->file_offset); | |
640 | f->file_offset = new_offset; | |
641 | } | |
642 | ||
53aa6171 | 643 | z = zbd_offset_to_zone(f, f->file_offset + f->io_size); |
0bf93a1a DLM |
644 | new_end = z->start; |
645 | if ((td->o.td_ddir != TD_DDIR_READ) && | |
646 | (f->file_offset + f->io_size != new_end)) { | |
647 | if (new_end <= f->file_offset) { | |
648 | log_info("%s: io_size must be at least one zone\n", | |
649 | f->file_name); | |
650 | return false; | |
651 | } | |
652 | log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n", | |
653 | f->file_name, f->io_size, | |
654 | new_end - f->file_offset); | |
655 | f->io_size = new_end - f->file_offset; | |
656 | } | |
657 | ||
658 | return true; | |
659 | } | |
660 | ||
bfbdd35b BVA |
661 | /* |
662 | * Verify whether offset and size parameters are aligned with zone boundaries. | |
663 | */ | |
664 | static bool zbd_verify_sizes(void) | |
665 | { | |
bfbdd35b | 666 | struct fio_file *f; |
da8f124f | 667 | int j; |
bfbdd35b | 668 | |
da8f124f | 669 | for_each_td(td) { |
bfbdd35b | 670 | for_each_file(td, f, j) { |
0bf93a1a | 671 | if (!zbd_zone_align_file_sizes(td, f)) |
4d37720a | 672 | return false; |
bfbdd35b | 673 | } |
da8f124f | 674 | } end_for_each(); |
bfbdd35b BVA |
675 | |
676 | return true; | |
677 | } | |
678 | ||
679 | static bool zbd_verify_bs(void) | |
680 | { | |
bfbdd35b | 681 | struct fio_file *f; |
da8f124f | 682 | int j; |
bfbdd35b | 683 | |
da8f124f | 684 | for_each_td(td) { |
e3be810b SK |
685 | if (td_trim(td) && |
686 | (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] || | |
687 | td->o.bssplit_nr[DDIR_TRIM])) { | |
688 | log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n"); | |
689 | return false; | |
690 | } | |
bfbdd35b | 691 | for_each_file(td, f, j) { |
1ddd225e AD |
692 | uint64_t zone_size; |
693 | ||
bfbdd35b BVA |
694 | if (!f->zbd_info) |
695 | continue; | |
139d8dc6 | 696 | |
bfbdd35b | 697 | zone_size = f->zbd_info->zone_size; |
e3be810b | 698 | if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) { |
ee5e3436 | 699 | log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n", |
e3be810b | 700 | f->file_name, td->o.bs[DDIR_TRIM], |
ee5e3436 | 701 | zone_size); |
e3be810b SK |
702 | return false; |
703 | } | |
bfbdd35b | 704 | } |
da8f124f | 705 | } end_for_each(); |
bfbdd35b BVA |
706 | return true; |
707 | } | |
708 | ||
bfbdd35b BVA |
709 | static int ilog2(uint64_t i) |
710 | { | |
711 | int log = -1; | |
712 | ||
713 | while (i) { | |
714 | i >>= 1; | |
715 | log++; | |
716 | } | |
717 | return log; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Initialize f->zbd_info for devices that are not zoned block devices. This | |
722 | * allows to execute a ZBD workload against a non-ZBD device. | |
723 | */ | |
724 | static int init_zone_info(struct thread_data *td, struct fio_file *f) | |
725 | { | |
726 | uint32_t nr_zones; | |
727 | struct fio_zone_info *p; | |
a4b7f12b | 728 | uint64_t zone_size = td->o.zone_size; |
b8dd9750 | 729 | uint64_t zone_capacity = td->o.zone_capacity; |
bfbdd35b | 730 | struct zoned_block_device_info *zbd_info = NULL; |
bfbdd35b BVA |
731 | int i; |
732 | ||
a4b7f12b | 733 | if (zone_size == 0) { |
9db0cde8 | 734 | log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", |
a4b7f12b DLM |
735 | f->file_name); |
736 | return 1; | |
737 | } | |
738 | ||
739 | if (zone_size < 512) { | |
740 | log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n", | |
741 | f->file_name); | |
742 | return 1; | |
743 | } | |
744 | ||
b8dd9750 HH |
745 | if (zone_capacity == 0) |
746 | zone_capacity = zone_size; | |
747 | ||
748 | if (zone_capacity > zone_size) { | |
749 | log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", | |
ee5e3436 | 750 | f->file_name, td->o.zone_capacity, td->o.zone_size); |
b8dd9750 HH |
751 | return 1; |
752 | } | |
753 | ||
9db0cde8 NC |
754 | if (f->real_file_size < zone_size) { |
755 | log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", | |
756 | f->file_name, f->real_file_size, zone_size); | |
757 | return -EINVAL; | |
758 | } | |
759 | ||
ee3696bd | 760 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
761 | zbd_info = scalloc(1, sizeof(*zbd_info) + |
762 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
763 | if (!zbd_info) | |
764 | return -ENOMEM; | |
765 | ||
44ec32cb | 766 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
767 | zbd_info->refcount = 1; |
768 | p = &zbd_info->zone_info[0]; | |
769 | for (i = 0; i < nr_zones; i++, p++) { | |
44ec32cb SK |
770 | mutex_init_pshared_with_type(&p->mutex, |
771 | PTHREAD_MUTEX_RECURSIVE); | |
bfbdd35b | 772 | p->start = i * zone_size; |
b14651a2 | 773 | p->wp = p->start; |
b7694961 DLM |
774 | p->type = ZBD_ZONE_TYPE_SWR; |
775 | p->cond = ZBD_ZONE_COND_EMPTY; | |
b8dd9750 | 776 | p->capacity = zone_capacity; |
be7a6bae | 777 | p->has_wp = 1; |
bfbdd35b BVA |
778 | } |
779 | /* a sentinel */ | |
780 | p->start = nr_zones * zone_size; | |
781 | ||
782 | f->zbd_info = zbd_info; | |
783 | f->zbd_info->zone_size = zone_size; | |
784 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 785 | ilog2(zone_size) : 0; |
bfbdd35b | 786 | f->zbd_info->nr_zones = nr_zones; |
bfbdd35b BVA |
787 | return 0; |
788 | } | |
789 | ||
790 | /* | |
b7694961 DLM |
791 | * Maximum number of zones to report in one operation. |
792 | */ | |
793 | #define ZBD_REPORT_MAX_ZONES 8192U | |
794 | ||
795 | /* | |
796 | * Parse the device zone report and store it in f->zbd_info. Must be called | |
797 | * only for devices that are zoned, namely those with a model != ZBD_NONE. | |
bfbdd35b BVA |
798 | */ |
799 | static int parse_zone_info(struct thread_data *td, struct fio_file *f) | |
800 | { | |
b7694961 DLM |
801 | int nr_zones, nrz; |
802 | struct zbd_zone *zones, *z; | |
bfbdd35b | 803 | struct fio_zone_info *p; |
04f9090b BVA |
804 | uint64_t zone_size, offset, capacity; |
805 | bool same_zone_cap = true; | |
bfbdd35b | 806 | struct zoned_block_device_info *zbd_info = NULL; |
d060babc | 807 | int i, j, ret = -ENOMEM; |
bfbdd35b | 808 | |
b7694961 DLM |
809 | zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); |
810 | if (!zones) | |
bfbdd35b BVA |
811 | goto out; |
812 | ||
b7694961 DLM |
813 | nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); |
814 | if (nrz < 0) { | |
815 | ret = nrz; | |
816 | log_info("fio: report zones (offset 0) failed for %s (%d).\n", | |
817 | f->file_name, -ret); | |
818 | goto out; | |
bfbdd35b BVA |
819 | } |
820 | ||
b7694961 | 821 | zone_size = zones[0].len; |
04f9090b | 822 | capacity = zones[0].capacity; |
ee3696bd | 823 | nr_zones = (f->real_file_size + zone_size - 1) / zone_size; |
bfbdd35b BVA |
824 | |
825 | if (td->o.zone_size == 0) { | |
ee3696bd DLM |
826 | td->o.zone_size = zone_size; |
827 | } else if (td->o.zone_size != zone_size) { | |
ee5e3436 SK |
828 | log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", |
829 | f->file_name, td->o.zone_size, zone_size); | |
bfbdd35b | 830 | ret = -EINVAL; |
b7694961 | 831 | goto out; |
bfbdd35b BVA |
832 | } |
833 | ||
9724b4f5 NC |
834 | dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", |
835 | f->file_name, nr_zones, zone_size / 1024); | |
bfbdd35b BVA |
836 | |
837 | zbd_info = scalloc(1, sizeof(*zbd_info) + | |
838 | (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); | |
bfbdd35b | 839 | if (!zbd_info) |
b7694961 | 840 | goto out; |
44ec32cb | 841 | mutex_init_pshared(&zbd_info->mutex); |
bfbdd35b BVA |
842 | zbd_info->refcount = 1; |
843 | p = &zbd_info->zone_info[0]; | |
b7694961 DLM |
844 | for (offset = 0, j = 0; j < nr_zones;) { |
845 | z = &zones[0]; | |
846 | for (i = 0; i < nrz; i++, j++, z++, p++) { | |
44ec32cb SK |
847 | mutex_init_pshared_with_type(&p->mutex, |
848 | PTHREAD_MUTEX_RECURSIVE); | |
b7694961 | 849 | p->start = z->start; |
236d23a8 | 850 | p->capacity = z->capacity; |
04f9090b BVA |
851 | if (capacity != z->capacity) |
852 | same_zone_cap = false; | |
139d8dc6 | 853 | |
bfbdd35b | 854 | switch (z->cond) { |
b7694961 DLM |
855 | case ZBD_ZONE_COND_NOT_WP: |
856 | case ZBD_ZONE_COND_FULL: | |
236d23a8 | 857 | p->wp = p->start + p->capacity; |
bfbdd35b BVA |
858 | break; |
859 | default: | |
860 | assert(z->start <= z->wp); | |
b7694961 DLM |
861 | assert(z->wp <= z->start + zone_size); |
862 | p->wp = z->wp; | |
bfbdd35b BVA |
863 | break; |
864 | } | |
be7a6bae DF |
865 | |
866 | switch (z->type) { | |
867 | case ZBD_ZONE_TYPE_SWR: | |
868 | p->has_wp = 1; | |
869 | break; | |
870 | default: | |
871 | p->has_wp = 0; | |
872 | } | |
bfbdd35b BVA |
873 | p->type = z->type; |
874 | p->cond = z->cond; | |
be7a6bae | 875 | |
bfbdd35b | 876 | if (j > 0 && p->start != p[-1].start + zone_size) { |
adfa7b7c BVA |
877 | log_info("%s: invalid zone data [%d:%d]: %"PRIu64" + %"PRIu64" != %"PRIu64"\n", |
878 | f->file_name, j, i, | |
879 | p[-1].start, zone_size, p->start); | |
bfbdd35b | 880 | ret = -EINVAL; |
b7694961 | 881 | goto out; |
bfbdd35b BVA |
882 | } |
883 | } | |
884 | z--; | |
b7694961 | 885 | offset = z->start + z->len; |
bfbdd35b BVA |
886 | if (j >= nr_zones) |
887 | break; | |
139d8dc6 | 888 | |
6c3f1cc1 DF |
889 | nrz = zbd_report_zones(td, f, offset, zones, |
890 | min((uint32_t)(nr_zones - j), | |
891 | ZBD_REPORT_MAX_ZONES)); | |
b7694961 DLM |
892 | if (nrz < 0) { |
893 | ret = nrz; | |
ee5e3436 SK |
894 | log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n", |
895 | offset, f->file_name, -ret); | |
b7694961 | 896 | goto out; |
bfbdd35b BVA |
897 | } |
898 | } | |
b7694961 | 899 | |
bfbdd35b | 900 | /* a sentinel */ |
b7694961 | 901 | zbd_info->zone_info[nr_zones].start = offset; |
bfbdd35b BVA |
902 | |
903 | f->zbd_info = zbd_info; | |
904 | f->zbd_info->zone_size = zone_size; | |
905 | f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? | |
ebc403fe | 906 | ilog2(zone_size) : 0; |
bfbdd35b | 907 | f->zbd_info->nr_zones = nr_zones; |
04f9090b BVA |
908 | |
909 | if (same_zone_cap) | |
910 | dprint(FD_ZBD, "Zone capacity = %"PRIu64" KB\n", | |
911 | capacity / 1024); | |
912 | ||
bfbdd35b BVA |
913 | zbd_info = NULL; |
914 | ret = 0; | |
915 | ||
bfbdd35b | 916 | out: |
b7694961 DLM |
917 | sfree(zbd_info); |
918 | free(zones); | |
bfbdd35b BVA |
919 | return ret; |
920 | } | |
921 | ||
a4807046 | 922 | static int zbd_set_max_write_zones(struct thread_data *td, struct fio_file *f) |
d2f442bc NC |
923 | { |
924 | struct zoned_block_device_info *zbd = f->zbd_info; | |
925 | unsigned int max_open_zones; | |
926 | int ret; | |
927 | ||
575686bb | 928 | if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { |
d2f442bc | 929 | /* Only host-managed devices have a max open limit */ |
a4807046 | 930 | zbd->max_write_zones = td->o.max_open_zones; |
d2f442bc NC |
931 | goto out; |
932 | } | |
933 | ||
934 | /* If host-managed, get the max open limit */ | |
935 | ret = zbd_get_max_open_zones(td, f, &max_open_zones); | |
936 | if (ret) | |
937 | return ret; | |
938 | ||
939 | if (!max_open_zones) { | |
940 | /* No device limit */ | |
a4807046 | 941 | zbd->max_write_zones = td->o.max_open_zones; |
d2f442bc NC |
942 | } else if (!td->o.max_open_zones) { |
943 | /* No user limit. Set limit to device limit */ | |
a4807046 | 944 | zbd->max_write_zones = max_open_zones; |
d2f442bc NC |
945 | } else if (td->o.max_open_zones <= max_open_zones) { |
946 | /* Both user limit and dev limit. User limit not too large */ | |
a4807046 | 947 | zbd->max_write_zones = td->o.max_open_zones; |
d2f442bc NC |
948 | } else { |
949 | /* Both user limit and dev limit. User limit too large */ | |
950 | td_verror(td, EINVAL, | |
951 | "Specified --max_open_zones is too large"); | |
952 | log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", | |
953 | td->o.max_open_zones, max_open_zones); | |
954 | return -EINVAL; | |
955 | } | |
956 | ||
957 | out: | |
958 | /* Ensure that the limit is not larger than FIO's internal limit */ | |
a4807046 | 959 | if (zbd->max_write_zones > ZBD_MAX_WRITE_ZONES) { |
b346af90 | 960 | td_verror(td, EINVAL, "'max_open_zones' value is too large"); |
139d8dc6 | 961 | log_err("'max_open_zones' value is larger than %u\n", |
a4807046 | 962 | ZBD_MAX_WRITE_ZONES); |
b346af90 NC |
963 | return -EINVAL; |
964 | } | |
965 | ||
a4807046 SK |
966 | dprint(FD_ZBD, "%s: using max write zones limit: %"PRIu32"\n", |
967 | f->file_name, zbd->max_write_zones); | |
d2f442bc NC |
968 | |
969 | return 0; | |
970 | } | |
971 | ||
bfbdd35b BVA |
972 | /* |
973 | * Allocate zone information and store it into f->zbd_info if zonemode=zbd. | |
974 | * | |
975 | * Returns 0 upon success and a negative error code upon failure. | |
976 | */ | |
379e5f09 | 977 | static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) |
bfbdd35b | 978 | { |
b7694961 DLM |
979 | enum zbd_zoned_model zbd_model; |
980 | int ret; | |
bfbdd35b BVA |
981 | |
982 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
983 | ||
b7694961 DLM |
984 | ret = zbd_get_zoned_model(td, f, &zbd_model); |
985 | if (ret) | |
986 | return ret; | |
987 | ||
bfbdd35b | 988 | switch (zbd_model) { |
b7694961 DLM |
989 | case ZBD_HOST_AWARE: |
990 | case ZBD_HOST_MANAGED: | |
bfbdd35b | 991 | ret = parse_zone_info(td, f); |
d2f442bc NC |
992 | if (ret) |
993 | return ret; | |
bfbdd35b | 994 | break; |
b7694961 | 995 | case ZBD_NONE: |
bfbdd35b | 996 | ret = init_zone_info(td, f); |
d2f442bc NC |
997 | if (ret) |
998 | return ret; | |
bfbdd35b | 999 | break; |
b7694961 DLM |
1000 | default: |
1001 | td_verror(td, EINVAL, "Unsupported zoned model"); | |
1002 | log_err("Unsupported zoned model\n"); | |
1003 | return -EINVAL; | |
bfbdd35b | 1004 | } |
b7694961 | 1005 | |
2c7dd23e | 1006 | assert(f->zbd_info); |
d2f442bc NC |
1007 | f->zbd_info->model = zbd_model; |
1008 | ||
a4807046 | 1009 | ret = zbd_set_max_write_zones(td, f); |
d2f442bc NC |
1010 | if (ret) { |
1011 | zbd_free_zone_info(f); | |
1012 | return ret; | |
219c662d | 1013 | } |
d2f442bc NC |
1014 | |
1015 | return 0; | |
bfbdd35b BVA |
1016 | } |
1017 | ||
1018 | void zbd_free_zone_info(struct fio_file *f) | |
1019 | { | |
1020 | uint32_t refcount; | |
1021 | ||
3c1dc34c | 1022 | assert(f->zbd_info); |
bfbdd35b BVA |
1023 | |
1024 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1025 | refcount = --f->zbd_info->refcount; | |
1026 | pthread_mutex_unlock(&f->zbd_info->mutex); | |
1027 | ||
1028 | assert((int32_t)refcount >= 0); | |
1029 | if (refcount == 0) | |
1030 | sfree(f->zbd_info); | |
1031 | f->zbd_info = NULL; | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * Initialize f->zbd_info. | |
1036 | * | |
1037 | * Returns 0 upon success and a negative error code upon failure. | |
1038 | * | |
1039 | * Note: this function can only work correctly if it is called before the first | |
1040 | * fio fork() call. | |
1041 | */ | |
1042 | static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) | |
1043 | { | |
bfbdd35b | 1044 | struct fio_file *f2; |
da8f124f | 1045 | int j, ret; |
bfbdd35b | 1046 | |
da8f124f | 1047 | for_each_td(td2) { |
bfbdd35b BVA |
1048 | for_each_file(td2, f2, j) { |
1049 | if (td2 == td && f2 == file) | |
1050 | continue; | |
1051 | if (!f2->zbd_info || | |
1052 | strcmp(f2->file_name, file->file_name) != 0) | |
1053 | continue; | |
1054 | file->zbd_info = f2->zbd_info; | |
1055 | file->zbd_info->refcount++; | |
1056 | return 0; | |
1057 | } | |
da8f124f | 1058 | } end_for_each(); |
bfbdd35b BVA |
1059 | |
1060 | ret = zbd_create_zone_info(td, file); | |
1061 | if (ret < 0) | |
c5837eec | 1062 | td_verror(td, -ret, "zbd_create_zone_info() failed"); |
139d8dc6 | 1063 | |
bfbdd35b BVA |
1064 | return ret; |
1065 | } | |
1066 | ||
8f39afa7 | 1067 | int zbd_init_files(struct thread_data *td) |
bfbdd35b BVA |
1068 | { |
1069 | struct fio_file *f; | |
1070 | int i; | |
1071 | ||
1072 | for_each_file(td, f, i) { | |
a4b7f12b | 1073 | if (zbd_init_zone_info(td, f)) |
bfbdd35b | 1074 | return 1; |
bfbdd35b | 1075 | } |
139d8dc6 | 1076 | |
8f39afa7 AD |
1077 | return 0; |
1078 | } | |
1079 | ||
1080 | void zbd_recalc_options_with_zone_granularity(struct thread_data *td) | |
1081 | { | |
1082 | struct fio_file *f; | |
1083 | int i; | |
1084 | ||
1085 | for_each_file(td, f, i) { | |
1086 | struct zoned_block_device_info *zbd = f->zbd_info; | |
139d8dc6 | 1087 | uint64_t zone_size; |
8f39afa7 | 1088 | |
139d8dc6 DLM |
1089 | /* zonemode=strided doesn't get per-file zone size. */ |
1090 | zone_size = zbd ? zbd->zone_size : td->o.zone_size; | |
8f39afa7 AD |
1091 | if (zone_size == 0) |
1092 | continue; | |
1093 | ||
139d8dc6 | 1094 | if (td->o.size_nz > 0) |
8f39afa7 | 1095 | td->o.size = td->o.size_nz * zone_size; |
139d8dc6 | 1096 | if (td->o.io_size_nz > 0) |
8f39afa7 | 1097 | td->o.io_size = td->o.io_size_nz * zone_size; |
139d8dc6 | 1098 | if (td->o.start_offset_nz > 0) |
8f39afa7 | 1099 | td->o.start_offset = td->o.start_offset_nz * zone_size; |
139d8dc6 DLM |
1100 | if (td->o.offset_increment_nz > 0) |
1101 | td->o.offset_increment = | |
1102 | td->o.offset_increment_nz * zone_size; | |
1103 | if (td->o.zone_skip_nz > 0) | |
8f39afa7 | 1104 | td->o.zone_skip = td->o.zone_skip_nz * zone_size; |
8f39afa7 AD |
1105 | } |
1106 | } | |
1107 | ||
9fb714da SK |
1108 | static uint64_t zbd_verify_and_set_vdb(struct thread_data *td, |
1109 | const struct fio_file *f) | |
1110 | { | |
1111 | struct fio_zone_info *zb, *ze, *z; | |
1112 | uint64_t wp_vdb = 0; | |
1113 | struct zoned_block_device_info *zbdi = f->zbd_info; | |
1114 | ||
1115 | assert(td->runstate < TD_RUNNING); | |
1116 | assert(zbdi); | |
1117 | ||
1118 | if (!accounting_vdb(td, f)) | |
1119 | return 0; | |
1120 | ||
1121 | /* | |
1122 | * Ensure that the I/O range includes one or more sequential zones so | |
1123 | * that f->min_zone and f->max_zone have different values. | |
1124 | */ | |
1125 | if (!zbd_is_seq_job(f)) | |
1126 | return 0; | |
1127 | ||
1128 | if (zbdi->write_min_zone != zbdi->write_max_zone) { | |
1129 | if (zbdi->write_min_zone != f->min_zone || | |
1130 | zbdi->write_max_zone != f->max_zone) { | |
1131 | td_verror(td, EINVAL, | |
1132 | "multi-jobs with different write ranges are " | |
1133 | "not supported with zone_reset_threshold"); | |
1134 | log_err("multi-jobs with different write ranges are " | |
1135 | "not supported with zone_reset_threshold\n"); | |
1136 | } | |
1137 | return 0; | |
1138 | } | |
1139 | ||
1140 | zbdi->write_min_zone = f->min_zone; | |
1141 | zbdi->write_max_zone = f->max_zone; | |
1142 | ||
1143 | zb = zbd_get_zone(f, f->min_zone); | |
1144 | ze = zbd_get_zone(f, f->max_zone); | |
1145 | for (z = zb; z < ze; z++) | |
1146 | if (z->has_wp) | |
1147 | wp_vdb += z->wp - z->start; | |
1148 | ||
1149 | zbdi->wp_valid_data_bytes = wp_vdb; | |
1150 | ||
1151 | return wp_vdb; | |
1152 | } | |
1153 | ||
8f39afa7 AD |
1154 | int zbd_setup_files(struct thread_data *td) |
1155 | { | |
1156 | struct fio_file *f; | |
1157 | int i; | |
bfbdd35b BVA |
1158 | |
1159 | if (!zbd_using_direct_io()) { | |
1160 | log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); | |
1161 | return 1; | |
1162 | } | |
1163 | ||
1164 | if (!zbd_verify_sizes()) | |
1165 | return 1; | |
1166 | ||
1167 | if (!zbd_verify_bs()) | |
1168 | return 1; | |
1169 | ||
6e2da06a SK |
1170 | if (td->o.experimental_verify) { |
1171 | log_err("zonemode=zbd does not support experimental verify\n"); | |
1172 | return 1; | |
1173 | } | |
1174 | ||
219c662d AD |
1175 | for_each_file(td, f, i) { |
1176 | struct zoned_block_device_info *zbd = f->zbd_info; | |
954217b9 SK |
1177 | struct fio_zone_info *z; |
1178 | int zi; | |
9fb714da | 1179 | uint64_t vdb; |
219c662d | 1180 | |
5ddf46d0 | 1181 | assert(zbd); |
219c662d | 1182 | |
dc8a3d62 DLM |
1183 | f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset); |
1184 | f->max_zone = | |
1185 | zbd_offset_to_zone_idx(f, f->file_offset + f->io_size); | |
f952800a | 1186 | |
9fb714da SK |
1187 | vdb = zbd_verify_and_set_vdb(td, f); |
1188 | ||
1189 | dprint(FD_ZBD, "%s(%s): valid data bytes = %" PRIu64 "\n", | |
1190 | __func__, f->file_name, vdb); | |
1191 | ||
f952800a SK |
1192 | /* |
1193 | * When all zones in the I/O range are conventional, io_size | |
1194 | * can be smaller than zone size, making min_zone the same | |
1195 | * as max_zone. This is why the assert below needs to be made | |
1196 | * conditional. | |
1197 | */ | |
1198 | if (zbd_is_seq_job(f)) | |
1199 | assert(f->min_zone < f->max_zone); | |
1200 | ||
219c662d | 1201 | if (td->o.max_open_zones > 0 && |
a4807046 | 1202 | zbd->max_write_zones != td->o.max_open_zones) { |
219c662d AD |
1203 | log_err("Different 'max_open_zones' values\n"); |
1204 | return 1; | |
1205 | } | |
b346af90 NC |
1206 | |
1207 | /* | |
1208 | * The per job max open zones limit cannot be used without a | |
1209 | * global max open zones limit. (As the tracking of open zones | |
1210 | * is disabled when there is no global max open zones limit.) | |
1211 | */ | |
a4807046 | 1212 | if (td->o.job_max_open_zones && !zbd->max_write_zones) { |
b346af90 | 1213 | log_err("'job_max_open_zones' cannot be used without a global open zones limit\n"); |
219c662d AD |
1214 | return 1; |
1215 | } | |
954217b9 | 1216 | |
ea51055c | 1217 | /* |
a4807046 | 1218 | * zbd->max_write_zones is the global limit shared for all jobs |
ea51055c NC |
1219 | * that target the same zoned block device. Force sync the per |
1220 | * thread global limit with the actual global limit. (The real | |
1221 | * per thread/job limit is stored in td->o.job_max_open_zones). | |
1222 | */ | |
a4807046 | 1223 | td->o.max_open_zones = zbd->max_write_zones; |
ea51055c | 1224 | |
954217b9 SK |
1225 | for (zi = f->min_zone; zi < f->max_zone; zi++) { |
1226 | z = &zbd->zone_info[zi]; | |
1227 | if (z->cond != ZBD_ZONE_COND_IMP_OPEN && | |
1228 | z->cond != ZBD_ZONE_COND_EXP_OPEN) | |
1229 | continue; | |
f539b98c | 1230 | if (__zbd_write_zone_get(td, f, z)) |
954217b9 SK |
1231 | continue; |
1232 | /* | |
1233 | * If the number of open zones exceeds specified limits, | |
8ac76889 | 1234 | * error out. |
954217b9 | 1235 | */ |
8ac76889 SK |
1236 | log_err("Number of open zones exceeds max_open_zones limit\n"); |
1237 | return 1; | |
954217b9 | 1238 | } |
219c662d AD |
1239 | } |
1240 | ||
bfbdd35b BVA |
1241 | return 0; |
1242 | } | |
1243 | ||
a7c2b6fc BVA |
1244 | /* |
1245 | * Reset zbd_info.write_cnt, the counter that counts down towards the next | |
1246 | * zone reset. | |
1247 | */ | |
1bb1bcad AD |
1248 | static void _zbd_reset_write_cnt(const struct thread_data *td, |
1249 | const struct fio_file *f) | |
a7c2b6fc BVA |
1250 | { |
1251 | assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); | |
1252 | ||
a7c2b6fc BVA |
1253 | f->zbd_info->write_cnt = td->o.zrf.u.f ? |
1254 | min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; | |
1bb1bcad AD |
1255 | } |
1256 | ||
1257 | static void zbd_reset_write_cnt(const struct thread_data *td, | |
1258 | const struct fio_file *f) | |
1259 | { | |
1260 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1261 | _zbd_reset_write_cnt(td, f); | |
a7c2b6fc BVA |
1262 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1263 | } | |
1264 | ||
1265 | static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, | |
1266 | const struct fio_file *f) | |
1267 | { | |
1268 | uint32_t write_cnt = 0; | |
1269 | ||
1270 | pthread_mutex_lock(&f->zbd_info->mutex); | |
1271 | assert(f->zbd_info->write_cnt); | |
1272 | if (f->zbd_info->write_cnt) | |
1273 | write_cnt = --f->zbd_info->write_cnt; | |
1274 | if (write_cnt == 0) | |
1bb1bcad | 1275 | _zbd_reset_write_cnt(td, f); |
a7c2b6fc BVA |
1276 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1277 | ||
1278 | return write_cnt == 0; | |
1279 | } | |
1280 | ||
bfbdd35b BVA |
1281 | void zbd_file_reset(struct thread_data *td, struct fio_file *f) |
1282 | { | |
91d25131 | 1283 | struct fio_zone_info *zb, *ze; |
c5c8b92b | 1284 | bool verify_data_left = false; |
bfbdd35b | 1285 | |
767d1372 | 1286 | if (!f->zbd_info || !td_write(td)) |
bfbdd35b BVA |
1287 | return; |
1288 | ||
39e06ee7 DLM |
1289 | zb = zbd_get_zone(f, f->min_zone); |
1290 | ze = zbd_get_zone(f, f->max_zone); | |
139d8dc6 | 1291 | |
bfbdd35b BVA |
1292 | /* |
1293 | * If data verification is enabled reset the affected zones before | |
1294 | * writing any data to avoid that a zone reset has to be issued while | |
1295 | * writing data, which causes data loss. | |
1296 | */ | |
c5c8b92b SK |
1297 | if (td->o.verify != VERIFY_NONE) { |
1298 | verify_data_left = td->runstate == TD_VERIFYING || | |
1299 | td->io_hist_len || td->verify_batch; | |
1300 | if (td->io_hist_len && td->o.verify_backlog) | |
1301 | verify_data_left = | |
1302 | td->io_hist_len % td->o.verify_backlog; | |
1303 | if (!verify_data_left) | |
1304 | zbd_reset_zones(td, f, zb, ze); | |
1305 | } | |
1306 | ||
a7c2b6fc | 1307 | zbd_reset_write_cnt(td, f); |
bfbdd35b BVA |
1308 | } |
1309 | ||
a4807046 | 1310 | /* Return random zone index for one of the write target zones. */ |
6463db6c AD |
1311 | static uint32_t pick_random_zone_idx(const struct fio_file *f, |
1312 | const struct io_u *io_u) | |
1313 | { | |
139d8dc6 | 1314 | return (io_u->offset - f->file_offset) * |
a4807046 | 1315 | f->zbd_info->num_write_zones / f->io_size; |
6463db6c AD |
1316 | } |
1317 | ||
0f77c977 SK |
1318 | static bool any_io_in_flight(void) |
1319 | { | |
da8f124f | 1320 | for_each_td(td) { |
0f77c977 SK |
1321 | if (td->io_u_in_flight) |
1322 | return true; | |
da8f124f | 1323 | } end_for_each(); |
0f77c977 SK |
1324 | |
1325 | return false; | |
1326 | } | |
1327 | ||
59b07544 | 1328 | /* |
a4807046 SK |
1329 | * Modify the offset of an I/O unit that does not refer to a zone such that |
1330 | * in write target zones array. Add a zone to or remove a zone from the lsit if | |
1331 | * necessary. The write target zone is searched across sequential zones. | |
21c0c884 | 1332 | * This algorithm can only work correctly if all write pointers are |
59b07544 BVA |
1333 | * a multiple of the fio block size. The caller must neither hold z->mutex |
1334 | * nor f->zbd_info->mutex. Returns with z->mutex held upon success. | |
1335 | */ | |
a4807046 SK |
1336 | static struct fio_zone_info *zbd_convert_to_write_zone(struct thread_data *td, |
1337 | struct io_u *io_u) | |
59b07544 | 1338 | { |
07fc3f57 | 1339 | const uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
fae3b9a0 | 1340 | struct fio_file *f = io_u->file; |
af94a8c3 | 1341 | struct zoned_block_device_info *zbdi = f->zbd_info; |
59b07544 | 1342 | struct fio_zone_info *z; |
a4807046 | 1343 | unsigned int write_zone_idx = -1; |
59b07544 BVA |
1344 | uint32_t zone_idx, new_zone_idx; |
1345 | int i; | |
a4807046 | 1346 | bool wait_zone_write; |
0f77c977 SK |
1347 | bool in_flight; |
1348 | bool should_retry = true; | |
59b07544 BVA |
1349 | |
1350 | assert(is_valid_offset(f, io_u->offset)); | |
1351 | ||
a4807046 | 1352 | if (zbdi->max_write_zones || td->o.job_max_open_zones) { |
59b07544 | 1353 | /* |
a4807046 | 1354 | * This statement accesses zbdi->write_zones[] on purpose |
59b07544 BVA |
1355 | * without locking. |
1356 | */ | |
a4807046 | 1357 | zone_idx = zbdi->write_zones[pick_random_zone_idx(f, io_u)]; |
59b07544 | 1358 | } else { |
dc8a3d62 | 1359 | zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); |
59b07544 | 1360 | } |
fae3b9a0 AD |
1361 | if (zone_idx < f->min_zone) |
1362 | zone_idx = f->min_zone; | |
1363 | else if (zone_idx >= f->max_zone) | |
1364 | zone_idx = f->max_zone - 1; | |
139d8dc6 DLM |
1365 | |
1366 | dprint(FD_ZBD, | |
1367 | "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", | |
59b07544 BVA |
1368 | __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); |
1369 | ||
1370 | /* | |
af94a8c3 | 1371 | * Since z->mutex is the outer lock and zbdi->mutex the inner |
59b07544 | 1372 | * lock it can happen that the state of the zone with index zone_idx |
af94a8c3 | 1373 | * has changed after 'z' has been assigned and before zbdi->mutex |
59b07544 BVA |
1374 | * has been obtained. Hence the loop. |
1375 | */ | |
1376 | for (;;) { | |
6463db6c AD |
1377 | uint32_t tmp_idx; |
1378 | ||
39e06ee7 | 1379 | z = zbd_get_zone(f, zone_idx); |
14351148 DF |
1380 | if (z->has_wp) |
1381 | zone_lock(td, f, z); | |
139d8dc6 | 1382 | |
af94a8c3 | 1383 | pthread_mutex_lock(&zbdi->mutex); |
139d8dc6 | 1384 | |
14351148 DF |
1385 | if (z->has_wp) { |
1386 | if (z->cond != ZBD_ZONE_COND_OFFLINE && | |
a4807046 | 1387 | zbdi->max_write_zones == 0 && |
139d8dc6 | 1388 | td->o.job_max_open_zones == 0) |
14351148 | 1389 | goto examine_zone; |
a4807046 SK |
1390 | if (zbdi->num_write_zones == 0) { |
1391 | dprint(FD_ZBD, "%s(%s): no zone is write target\n", | |
14351148 | 1392 | __func__, f->file_name); |
a4807046 | 1393 | goto choose_other_zone; |
14351148 | 1394 | } |
59b07544 | 1395 | } |
6463db6c AD |
1396 | |
1397 | /* | |
a4807046 | 1398 | * Array of write target zones is per-device, shared across all |
139d8dc6 DLM |
1399 | * threads. Start with quasi-random candidate zone. Ignore |
1400 | * zones which don't belong to thread's offset/size area. | |
6463db6c | 1401 | */ |
a4807046 SK |
1402 | write_zone_idx = pick_random_zone_idx(f, io_u); |
1403 | assert(!write_zone_idx || | |
1404 | write_zone_idx < zbdi->num_write_zones); | |
1405 | tmp_idx = write_zone_idx; | |
139d8dc6 | 1406 | |
a4807046 | 1407 | for (i = 0; i < zbdi->num_write_zones; i++) { |
6463db6c AD |
1408 | uint32_t tmpz; |
1409 | ||
a4807046 | 1410 | if (tmp_idx >= zbdi->num_write_zones) |
6463db6c | 1411 | tmp_idx = 0; |
a4807046 | 1412 | tmpz = zbdi->write_zones[tmp_idx]; |
fae3b9a0 | 1413 | if (f->min_zone <= tmpz && tmpz < f->max_zone) { |
a4807046 | 1414 | write_zone_idx = tmp_idx; |
6463db6c AD |
1415 | goto found_candidate_zone; |
1416 | } | |
1417 | ||
1418 | tmp_idx++; | |
1419 | } | |
1420 | ||
1421 | dprint(FD_ZBD, "%s(%s): no candidate zone\n", | |
1422 | __func__, f->file_name); | |
139d8dc6 | 1423 | |
af94a8c3 | 1424 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1425 | |
14351148 DF |
1426 | if (z->has_wp) |
1427 | zone_unlock(z); | |
139d8dc6 | 1428 | |
6463db6c AD |
1429 | return NULL; |
1430 | ||
1431 | found_candidate_zone: | |
a4807046 | 1432 | new_zone_idx = zbdi->write_zones[write_zone_idx]; |
59b07544 BVA |
1433 | if (new_zone_idx == zone_idx) |
1434 | break; | |
1435 | zone_idx = new_zone_idx; | |
139d8dc6 | 1436 | |
af94a8c3 | 1437 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1438 | |
14351148 DF |
1439 | if (z->has_wp) |
1440 | zone_unlock(z); | |
59b07544 BVA |
1441 | } |
1442 | ||
af94a8c3 | 1443 | /* Both z->mutex and zbdi->mutex are held. */ |
59b07544 BVA |
1444 | |
1445 | examine_zone: | |
df67bf1e | 1446 | if (zbd_zone_remainder(z) >= min_bs) { |
af94a8c3 | 1447 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1448 | goto out; |
1449 | } | |
b2da58c4 | 1450 | |
a4807046 SK |
1451 | choose_other_zone: |
1452 | /* Check if number of write target zones reaches one of limits. */ | |
1453 | wait_zone_write = | |
1454 | zbdi->num_write_zones == f->max_zone - f->min_zone || | |
1455 | (zbdi->max_write_zones && | |
1456 | zbdi->num_write_zones == zbdi->max_write_zones) || | |
b2da58c4 | 1457 | (td->o.job_max_open_zones && |
a4807046 | 1458 | td->num_write_zones == td->o.job_max_open_zones); |
b2da58c4 | 1459 | |
af94a8c3 | 1460 | pthread_mutex_unlock(&zbdi->mutex); |
59b07544 BVA |
1461 | |
1462 | /* Only z->mutex is held. */ | |
1463 | ||
b2da58c4 | 1464 | /* |
a4807046 SK |
1465 | * When number of write target zones reaches to one of limits, wait for |
1466 | * zone write completion to one of them before trying a new zone. | |
b2da58c4 | 1467 | */ |
a4807046 | 1468 | if (wait_zone_write) { |
139d8dc6 | 1469 | dprint(FD_ZBD, |
a4807046 | 1470 | "%s(%s): quiesce to remove a zone from write target zones array\n", |
b2da58c4 SK |
1471 | __func__, f->file_name); |
1472 | io_u_quiesce(td); | |
1473 | } | |
1474 | ||
0f77c977 | 1475 | retry: |
a4807046 | 1476 | /* Zone 'z' is full, so try to choose a new zone. */ |
af94a8c3 | 1477 | for (i = f->io_size / zbdi->zone_size; i > 0; i--) { |
59b07544 | 1478 | zone_idx++; |
21c0c884 SK |
1479 | if (z->has_wp) |
1480 | zone_unlock(z); | |
59b07544 | 1481 | z++; |
ee3696bd | 1482 | if (!is_valid_offset(f, z->start)) { |
59b07544 | 1483 | /* Wrap-around. */ |
fae3b9a0 | 1484 | zone_idx = f->min_zone; |
39e06ee7 | 1485 | z = zbd_get_zone(f, zone_idx); |
59b07544 | 1486 | } |
ee3696bd | 1487 | assert(is_valid_offset(f, z->start)); |
21c0c884 SK |
1488 | if (!z->has_wp) |
1489 | continue; | |
fae3b9a0 | 1490 | zone_lock(td, f, z); |
a4807046 | 1491 | if (z->write) |
59b07544 | 1492 | continue; |
a4807046 | 1493 | if (zbd_write_zone_get(td, f, z)) |
59b07544 BVA |
1494 | goto out; |
1495 | } | |
1496 | ||
1497 | /* Only z->mutex is held. */ | |
1498 | ||
a4807046 | 1499 | /* Check whether the write fits in any of the write target zones. */ |
af94a8c3 | 1500 | pthread_mutex_lock(&zbdi->mutex); |
a4807046 SK |
1501 | for (i = 0; i < zbdi->num_write_zones; i++) { |
1502 | zone_idx = zbdi->write_zones[i]; | |
fae3b9a0 AD |
1503 | if (zone_idx < f->min_zone || zone_idx >= f->max_zone) |
1504 | continue; | |
af94a8c3 | 1505 | pthread_mutex_unlock(&zbdi->mutex); |
4d4c71e6 | 1506 | zone_unlock(z); |
59b07544 | 1507 | |
39e06ee7 | 1508 | z = zbd_get_zone(f, zone_idx); |
59b07544 | 1509 | |
fae3b9a0 | 1510 | zone_lock(td, f, z); |
df67bf1e | 1511 | if (zbd_zone_remainder(z) >= min_bs) |
59b07544 | 1512 | goto out; |
af94a8c3 | 1513 | pthread_mutex_lock(&zbdi->mutex); |
59b07544 | 1514 | } |
0f77c977 SK |
1515 | |
1516 | /* | |
1517 | * When any I/O is in-flight or when all I/Os in-flight get completed, | |
a4807046 SK |
1518 | * the I/Os might have removed zones from the write target array then |
1519 | * retry the steps to choose a zone. Before retry, call io_u_quiesce() | |
1520 | * to complete in-flight writes. | |
0f77c977 SK |
1521 | */ |
1522 | in_flight = any_io_in_flight(); | |
1523 | if (in_flight || should_retry) { | |
139d8dc6 | 1524 | dprint(FD_ZBD, |
a4807046 | 1525 | "%s(%s): wait zone write and retry write target zone selection\n", |
0f77c977 SK |
1526 | __func__, f->file_name); |
1527 | pthread_mutex_unlock(&zbdi->mutex); | |
1528 | zone_unlock(z); | |
1529 | io_u_quiesce(td); | |
1530 | zone_lock(td, f, z); | |
1531 | should_retry = in_flight; | |
1532 | goto retry; | |
1533 | } | |
1534 | ||
af94a8c3 | 1535 | pthread_mutex_unlock(&zbdi->mutex); |
139d8dc6 | 1536 | |
4d4c71e6 | 1537 | zone_unlock(z); |
139d8dc6 | 1538 | |
a4807046 | 1539 | dprint(FD_ZBD, "%s(%s): did not choose another write zone\n", |
139d8dc6 DLM |
1540 | __func__, f->file_name); |
1541 | ||
59b07544 BVA |
1542 | return NULL; |
1543 | ||
1544 | out: | |
139d8dc6 DLM |
1545 | dprint(FD_ZBD, "%s(%s): returning zone %d\n", |
1546 | __func__, f->file_name, zone_idx); | |
1547 | ||
ee3696bd | 1548 | io_u->offset = z->start; |
21c0c884 | 1549 | assert(z->has_wp); |
8a866de7 | 1550 | assert(z->cond != ZBD_ZONE_COND_OFFLINE); |
139d8dc6 | 1551 | |
59b07544 BVA |
1552 | return z; |
1553 | } | |
1554 | ||
bfbdd35b | 1555 | /* |
5c86fdf6 SK |
1556 | * Find another zone which has @min_bytes of readable data. Search in zones |
1557 | * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf. | |
bfbdd35b | 1558 | * |
21c0c884 SK |
1559 | * Either returns NULL or returns a zone pointer. When the zone has write |
1560 | * pointer, hold the mutex for the zone. | |
bfbdd35b BVA |
1561 | */ |
1562 | static struct fio_zone_info * | |
07fc3f57 | 1563 | zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes, |
bfbdd35b BVA |
1564 | struct fio_zone_info *zb, struct fio_zone_info *zl) |
1565 | { | |
fae3b9a0 | 1566 | struct fio_file *f = io_u->file; |
bfbdd35b | 1567 | struct fio_zone_info *z1, *z2; |
39e06ee7 | 1568 | const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone); |
bfbdd35b BVA |
1569 | |
1570 | /* | |
1571 | * Skip to the next non-empty zone in case of sequential I/O and to | |
1572 | * the nearest non-empty zone in case of random I/O. | |
1573 | */ | |
1574 | for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { | |
b7694961 | 1575 | if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1576 | if (z1->has_wp) |
1577 | zone_lock(td, f, z1); | |
5c86fdf6 | 1578 | if (z1->start + min_bytes <= z1->wp) |
bfbdd35b | 1579 | return z1; |
21c0c884 SK |
1580 | if (z1->has_wp) |
1581 | zone_unlock(z1); | |
bfbdd35b BVA |
1582 | } else if (!td_random(td)) { |
1583 | break; | |
1584 | } | |
139d8dc6 | 1585 | |
bfbdd35b | 1586 | if (td_random(td) && z2 >= zf && |
b7694961 | 1587 | z2->cond != ZBD_ZONE_COND_OFFLINE) { |
21c0c884 SK |
1588 | if (z2->has_wp) |
1589 | zone_lock(td, f, z2); | |
5c86fdf6 | 1590 | if (z2->start + min_bytes <= z2->wp) |
bfbdd35b | 1591 | return z2; |
21c0c884 SK |
1592 | if (z2->has_wp) |
1593 | zone_unlock(z2); | |
bfbdd35b BVA |
1594 | } |
1595 | } | |
139d8dc6 DLM |
1596 | |
1597 | dprint(FD_ZBD, | |
1598 | "%s: no zone has %"PRIu64" bytes of readable data\n", | |
5c86fdf6 | 1599 | f->file_name, min_bytes); |
139d8dc6 | 1600 | |
bfbdd35b BVA |
1601 | return NULL; |
1602 | } | |
1603 | ||
b2da58c4 SK |
1604 | /** |
1605 | * zbd_end_zone_io - update zone status at command completion | |
1606 | * @io_u: I/O unit | |
1607 | * @z: zone info pointer | |
1608 | * | |
a4807046 SK |
1609 | * If the write command made the zone full, remove it from the write target |
1610 | * zones array. | |
b2da58c4 SK |
1611 | * |
1612 | * The caller must hold z->mutex. | |
1613 | */ | |
1614 | static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, | |
1615 | struct fio_zone_info *z) | |
1616 | { | |
1617 | const struct fio_file *f = io_u->file; | |
1618 | ||
1619 | if (io_u->ddir == DDIR_WRITE && | |
1620 | io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { | |
1621 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a4807046 | 1622 | zbd_write_zone_put(td, f, z); |
b2da58c4 SK |
1623 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1624 | } | |
1625 | } | |
1626 | ||
bfbdd35b | 1627 | /** |
d9ed3e63 | 1628 | * zbd_queue_io - update the write pointer of a sequential zone |
bfbdd35b | 1629 | * @io_u: I/O unit |
d9ed3e63 DLM |
1630 | * @success: Whether or not the I/O unit has been queued successfully |
1631 | * @q: queueing status (busy, completed or queued). | |
bfbdd35b | 1632 | * |
d9ed3e63 DLM |
1633 | * For write and trim operations, update the write pointer of the I/O unit |
1634 | * target zone. | |
bfbdd35b | 1635 | */ |
b2da58c4 SK |
1636 | static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, |
1637 | bool success) | |
bfbdd35b | 1638 | { |
d9ed3e63 DLM |
1639 | const struct fio_file *f = io_u->file; |
1640 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
bfbdd35b | 1641 | struct fio_zone_info *z; |
d9ed3e63 | 1642 | uint64_t zone_end; |
bfbdd35b | 1643 | |
5ddf46d0 | 1644 | assert(zbd_info); |
bfbdd35b | 1645 | |
53aa6171 | 1646 | z = zbd_offset_to_zone(f, io_u->offset); |
43bcbd5b | 1647 | assert(z->has_wp); |
d9ed3e63 | 1648 | |
bfbdd35b BVA |
1649 | if (!success) |
1650 | goto unlock; | |
d9ed3e63 DLM |
1651 | |
1652 | dprint(FD_ZBD, | |
1653 | "%s: queued I/O (%lld, %llu) for zone %u\n", | |
53aa6171 | 1654 | f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z)); |
d9ed3e63 | 1655 | |
bfbdd35b BVA |
1656 | switch (io_u->ddir) { |
1657 | case DDIR_WRITE: | |
d9ed3e63 | 1658 | zone_end = min((uint64_t)(io_u->offset + io_u->buflen), |
236d23a8 | 1659 | zbd_zone_capacity_end(z)); |
139d8dc6 | 1660 | |
a7c2b6fc BVA |
1661 | /* |
1662 | * z->wp > zone_end means that one or more I/O errors | |
1663 | * have occurred. | |
1664 | */ | |
2fb29f27 SK |
1665 | if (accounting_vdb(td, f) && z->wp <= zone_end) { |
1666 | pthread_mutex_lock(&zbd_info->mutex); | |
d56a6df3 | 1667 | zbd_info->wp_valid_data_bytes += zone_end - z->wp; |
2fb29f27 SK |
1668 | pthread_mutex_unlock(&zbd_info->mutex); |
1669 | } | |
bfbdd35b BVA |
1670 | z->wp = zone_end; |
1671 | break; | |
bfbdd35b BVA |
1672 | default: |
1673 | break; | |
1674 | } | |
d9ed3e63 | 1675 | |
b2da58c4 SK |
1676 | if (q == FIO_Q_COMPLETED && !io_u->error) |
1677 | zbd_end_zone_io(td, io_u, z); | |
1678 | ||
bfbdd35b | 1679 | unlock: |
d9ed3e63 DLM |
1680 | if (!success || q != FIO_Q_QUEUED) { |
1681 | /* BUSY or COMPLETED: unlock the zone */ | |
4d4c71e6 | 1682 | zone_unlock(z); |
d9ed3e63 DLM |
1683 | io_u->zbd_put_io = NULL; |
1684 | } | |
1685 | } | |
1686 | ||
1687 | /** | |
1688 | * zbd_put_io - Unlock an I/O unit target zone lock | |
1689 | * @io_u: I/O unit | |
1690 | */ | |
b2da58c4 | 1691 | static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) |
d9ed3e63 DLM |
1692 | { |
1693 | const struct fio_file *f = io_u->file; | |
1694 | struct zoned_block_device_info *zbd_info = f->zbd_info; | |
1695 | struct fio_zone_info *z; | |
d9ed3e63 | 1696 | |
5ddf46d0 | 1697 | assert(zbd_info); |
615555bb | 1698 | |
53aa6171 | 1699 | z = zbd_offset_to_zone(f, io_u->offset); |
43bcbd5b | 1700 | assert(z->has_wp); |
d9ed3e63 DLM |
1701 | |
1702 | dprint(FD_ZBD, | |
1703 | "%s: terminate I/O (%lld, %llu) for zone %u\n", | |
53aa6171 | 1704 | f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z)); |
d9ed3e63 | 1705 | |
b2da58c4 SK |
1706 | zbd_end_zone_io(td, io_u, z); |
1707 | ||
4d4c71e6 | 1708 | zone_unlock(z); |
bfbdd35b BVA |
1709 | } |
1710 | ||
9d87c646 DLM |
1711 | /* |
1712 | * Windows and MacOS do not define this. | |
1713 | */ | |
1714 | #ifndef EREMOTEIO | |
1715 | #define EREMOTEIO 121 /* POSIX value */ | |
1716 | #endif | |
1717 | ||
bfbdd35b BVA |
1718 | bool zbd_unaligned_write(int error_code) |
1719 | { | |
1720 | switch (error_code) { | |
1721 | case EIO: | |
1722 | case EREMOTEIO: | |
1723 | return true; | |
1724 | } | |
1725 | return false; | |
1726 | } | |
1727 | ||
4d37720a DLM |
1728 | /** |
1729 | * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives | |
1730 | * @td: FIO thread data. | |
1731 | * @io_u: FIO I/O unit. | |
1732 | * | |
1733 | * For sequential workloads, change the file offset to skip zoneskip bytes when | |
1734 | * no more IO can be performed in the current zone. | |
1735 | * - For read workloads, zoneskip is applied when the io has reached the end of | |
1736 | * the zone or the zone write position (when td->o.read_beyond_wp is false). | |
1737 | * - For write workloads, zoneskip is applied when the zone is full. | |
1738 | * This applies only to read and write operations. | |
1739 | */ | |
1740 | void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) | |
1741 | { | |
1742 | struct fio_file *f = io_u->file; | |
1743 | enum fio_ddir ddir = io_u->ddir; | |
1744 | struct fio_zone_info *z; | |
4d37720a DLM |
1745 | |
1746 | assert(td->o.zone_mode == ZONE_MODE_ZBD); | |
1747 | assert(td->o.zone_size); | |
5ddf46d0 | 1748 | assert(f->zbd_info); |
4d37720a | 1749 | |
53aa6171 | 1750 | z = zbd_offset_to_zone(f, f->last_pos[ddir]); |
236d23a8 SK |
1751 | |
1752 | /* | |
1753 | * When the zone capacity is smaller than the zone size and the I/O is | |
1754 | * sequential write, skip to zone end if the latest position is at the | |
1755 | * zone capacity limit. | |
1756 | */ | |
139d8dc6 DLM |
1757 | if (z->capacity < f->zbd_info->zone_size && |
1758 | !td_random(td) && ddir == DDIR_WRITE && | |
236d23a8 SK |
1759 | f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { |
1760 | dprint(FD_ZBD, | |
1761 | "%s: Jump from zone capacity limit to zone end:" | |
ee5e3436 SK |
1762 | " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n", |
1763 | f->file_name, f->last_pos[ddir], | |
53aa6171 | 1764 | zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity); |
236d23a8 SK |
1765 | td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; |
1766 | f->last_pos[ddir] = zbd_zone_end(z); | |
1767 | } | |
1768 | ||
4d37720a DLM |
1769 | /* |
1770 | * zone_skip is valid only for sequential workloads. | |
1771 | */ | |
1772 | if (td_random(td) || !td->o.zone_skip) | |
1773 | return; | |
1774 | ||
1775 | /* | |
1776 | * It is time to switch to a new zone if: | |
1777 | * - zone_bytes == zone_size bytes have already been accessed | |
1778 | * - The last position reached the end of the current zone. | |
1779 | * - For reads with td->o.read_beyond_wp == false, the last position | |
1780 | * reached the zone write pointer. | |
1781 | */ | |
4d37720a | 1782 | if (td->zone_bytes >= td->o.zone_size || |
236d23a8 | 1783 | f->last_pos[ddir] >= zbd_zone_end(z) || |
4d37720a DLM |
1784 | (ddir == DDIR_READ && |
1785 | (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { | |
1786 | /* | |
1787 | * Skip zones. | |
1788 | */ | |
1789 | td->zone_bytes = 0; | |
1790 | f->file_offset += td->o.zone_size + td->o.zone_skip; | |
1791 | ||
1792 | /* | |
1793 | * Wrap from the beginning, if we exceed the file size | |
1794 | */ | |
1795 | if (f->file_offset >= f->real_file_size) | |
1796 | f->file_offset = get_start_offset(td, f); | |
1797 | ||
1798 | f->last_pos[ddir] = f->file_offset; | |
1799 | td->io_skip_bytes += td->o.zone_skip; | |
1800 | } | |
1801 | } | |
1802 | ||
c65057f9 | 1803 | /** |
c7d5e152 | 1804 | * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. |
c65057f9 SK |
1805 | * |
1806 | * @td: FIO thread data. | |
1807 | * @io_u: FIO I/O unit. | |
1808 | * @ddir: I/O direction before adjustment. | |
1809 | * | |
1810 | * Return adjusted I/O direction. | |
1811 | */ | |
1812 | enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, | |
1813 | enum fio_ddir ddir) | |
1814 | { | |
1815 | /* | |
1816 | * In case read direction is chosen for the first random I/O, fio with | |
1817 | * zonemode=zbd stops because no data can be read from zoned block | |
1818 | * devices with all empty zones. Overwrite the first I/O direction as | |
1819 | * write to make sure data to read exists. | |
1820 | */ | |
5ddf46d0 | 1821 | assert(io_u->file->zbd_info); |
731461cc | 1822 | if (ddir != DDIR_READ || !td_rw(td)) |
c65057f9 SK |
1823 | return ddir; |
1824 | ||
440f63b1 | 1825 | if (io_u->file->last_start[DDIR_WRITE] != -1ULL || td->o.read_beyond_wp) |
c65057f9 SK |
1826 | return DDIR_READ; |
1827 | ||
1828 | return DDIR_WRITE; | |
1829 | } | |
1830 | ||
bfbdd35b BVA |
1831 | /** |
1832 | * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives | |
1833 | * @td: FIO thread data. | |
1834 | * @io_u: FIO I/O unit. | |
1835 | * | |
1836 | * Locking strategy: returns with z->mutex locked if and only if z refers | |
1837 | * to a sequential zone and if io_u_accept is returned. z is the zone that | |
1838 | * corresponds to io_u->offset at the end of this function. | |
1839 | */ | |
1840 | enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) | |
1841 | { | |
b7694961 | 1842 | struct fio_file *f = io_u->file; |
af94a8c3 | 1843 | struct zoned_block_device_info *zbdi = f->zbd_info; |
de65f7b7 | 1844 | struct fio_zone_info *zb, *zl, *orig_zb; |
bfbdd35b | 1845 | uint32_t orig_len = io_u->buflen; |
07fc3f57 | 1846 | uint64_t min_bs = td->o.min_bs[io_u->ddir]; |
bfbdd35b BVA |
1847 | uint64_t new_len; |
1848 | int64_t range; | |
1849 | ||
af94a8c3 | 1850 | assert(zbdi); |
adc6adcb | 1851 | assert(min_bs); |
bfbdd35b BVA |
1852 | assert(is_valid_offset(f, io_u->offset)); |
1853 | assert(io_u->buflen); | |
139d8dc6 | 1854 | |
53aa6171 | 1855 | zb = zbd_offset_to_zone(f, io_u->offset); |
de65f7b7 | 1856 | orig_zb = zb; |
bfbdd35b | 1857 | |
2efcf74b SK |
1858 | if (!zb->has_wp) { |
1859 | /* Accept non-write I/Os for conventional zones. */ | |
1860 | if (io_u->ddir != DDIR_WRITE) | |
1861 | return io_u_accept; | |
139d8dc6 | 1862 | |
2efcf74b SK |
1863 | /* |
1864 | * Make sure that writes to conventional zones | |
1865 | * don't cross over to any sequential zones. | |
1866 | */ | |
1867 | if (!(zb + 1)->has_wp || | |
1868 | io_u->offset + io_u->buflen <= (zb + 1)->start) | |
1869 | return io_u_accept; | |
1870 | ||
1871 | if (io_u->offset + min_bs > (zb + 1)->start) { | |
1872 | dprint(FD_IO, | |
07fc3f57 | 1873 | "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n", |
1e3d6e03 | 1874 | f->file_name, io_u->offset, |
ee5e3436 | 1875 | min_bs, (zb + 1)->start); |
139d8dc6 DLM |
1876 | io_u->offset = |
1877 | zb->start + (zb + 1)->start - io_u->offset; | |
1878 | new_len = min(io_u->buflen, | |
1879 | (zb + 1)->start - io_u->offset); | |
2efcf74b SK |
1880 | } else { |
1881 | new_len = (zb + 1)->start - io_u->offset; | |
1882 | } | |
139d8dc6 | 1883 | |
2efcf74b | 1884 | io_u->buflen = new_len / min_bs * min_bs; |
139d8dc6 | 1885 | |
bfbdd35b | 1886 | return io_u_accept; |
2efcf74b | 1887 | } |
bfbdd35b BVA |
1888 | |
1889 | /* | |
1890 | * Accept the I/O offset for reads if reading beyond the write pointer | |
1891 | * is enabled. | |
1892 | */ | |
b7694961 | 1893 | if (zb->cond != ZBD_ZONE_COND_OFFLINE && |
bfbdd35b BVA |
1894 | io_u->ddir == DDIR_READ && td->o.read_beyond_wp) |
1895 | return io_u_accept; | |
1896 | ||
fae3b9a0 | 1897 | zone_lock(td, f, zb); |
6f0c6085 | 1898 | |
bfbdd35b BVA |
1899 | switch (io_u->ddir) { |
1900 | case DDIR_READ: | |
6e2da06a | 1901 | if (td->runstate == TD_VERIFYING && td_write(td)) |
bfbdd35b | 1902 | goto accept; |
139d8dc6 | 1903 | |
bfbdd35b | 1904 | /* |
de65f7b7 DLM |
1905 | * Check that there is enough written data in the zone to do an |
1906 | * I/O of at least min_bs B. If there isn't, find a new zone for | |
1907 | * the I/O. | |
bfbdd35b | 1908 | */ |
b7694961 | 1909 | range = zb->cond != ZBD_ZONE_COND_OFFLINE ? |
ee3696bd | 1910 | zb->wp - zb->start : 0; |
de65f7b7 | 1911 | if (range < min_bs || |
ee3696bd | 1912 | ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { |
4d4c71e6 | 1913 | zone_unlock(zb); |
39e06ee7 | 1914 | zl = zbd_get_zone(f, f->max_zone); |
5c86fdf6 | 1915 | zb = zbd_find_zone(td, io_u, min_bs, zb, zl); |
bfbdd35b BVA |
1916 | if (!zb) { |
1917 | dprint(FD_ZBD, | |
1918 | "%s: zbd_find_zone(%lld, %llu) failed\n", | |
1919 | f->file_name, io_u->offset, | |
1920 | io_u->buflen); | |
1921 | goto eof; | |
1922 | } | |
de65f7b7 DLM |
1923 | /* |
1924 | * zbd_find_zone() returned a zone with a range of at | |
1925 | * least min_bs. | |
1926 | */ | |
ee3696bd | 1927 | range = zb->wp - zb->start; |
de65f7b7 DLM |
1928 | assert(range >= min_bs); |
1929 | ||
1930 | if (!td_random(td)) | |
ee3696bd | 1931 | io_u->offset = zb->start; |
bfbdd35b | 1932 | } |
139d8dc6 | 1933 | |
de65f7b7 DLM |
1934 | /* |
1935 | * Make sure the I/O is within the zone valid data range while | |
1936 | * maximizing the I/O size and preserving randomness. | |
1937 | */ | |
1938 | if (range <= io_u->buflen) | |
ee3696bd | 1939 | io_u->offset = zb->start; |
de65f7b7 | 1940 | else if (td_random(td)) |
ee3696bd DLM |
1941 | io_u->offset = zb->start + |
1942 | ((io_u->offset - orig_zb->start) % | |
de65f7b7 | 1943 | (range - io_u->buflen)) / min_bs * min_bs; |
139d8dc6 | 1944 | |
43bcbd5b SK |
1945 | /* |
1946 | * When zbd_find_zone() returns a conventional zone, | |
1947 | * we can simply accept the new i/o offset here. | |
1948 | */ | |
1949 | if (!zb->has_wp) | |
1950 | return io_u_accept; | |
139d8dc6 | 1951 | |
de65f7b7 DLM |
1952 | /* |
1953 | * Make sure the I/O does not cross over the zone wp position. | |
1954 | */ | |
1955 | new_len = min((unsigned long long)io_u->buflen, | |
ee3696bd | 1956 | (unsigned long long)(zb->wp - io_u->offset)); |
de65f7b7 DLM |
1957 | new_len = new_len / min_bs * min_bs; |
1958 | if (new_len < io_u->buflen) { | |
1959 | io_u->buflen = new_len; | |
1960 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
1961 | orig_len, io_u->buflen); | |
bfbdd35b | 1962 | } |
139d8dc6 | 1963 | |
ee3696bd DLM |
1964 | assert(zb->start <= io_u->offset); |
1965 | assert(io_u->offset + io_u->buflen <= zb->wp); | |
139d8dc6 | 1966 | |
bfbdd35b | 1967 | goto accept; |
139d8dc6 | 1968 | |
bfbdd35b | 1969 | case DDIR_WRITE: |
af94a8c3 | 1970 | if (io_u->buflen > zbdi->zone_size) { |
1c74aadc DF |
1971 | td_verror(td, EINVAL, "I/O buflen exceeds zone size"); |
1972 | dprint(FD_IO, | |
ee5e3436 SK |
1973 | "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n", |
1974 | f->file_name, io_u->buflen, zbdi->zone_size); | |
bfbdd35b | 1975 | goto eof; |
1c74aadc | 1976 | } |
139d8dc6 | 1977 | |
e1a1b59b SK |
1978 | retry: |
1979 | if (zbd_zone_remainder(zb) > 0 && | |
1980 | zbd_zone_remainder(zb) < min_bs) { | |
1981 | pthread_mutex_lock(&f->zbd_info->mutex); | |
a4807046 | 1982 | zbd_write_zone_put(td, f, zb); |
e1a1b59b SK |
1983 | pthread_mutex_unlock(&f->zbd_info->mutex); |
1984 | dprint(FD_ZBD, | |
1985 | "%s: finish zone %d\n", | |
1986 | f->file_name, zbd_zone_idx(f, zb)); | |
1987 | io_u_quiesce(td); | |
1988 | zbd_finish_zone(td, f, zb); | |
1989 | if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) { | |
1990 | if (!td_random(td)) | |
1991 | goto eof; | |
1992 | } | |
1993 | zone_unlock(zb); | |
1994 | ||
1995 | /* Find the next write pointer zone */ | |
1996 | do { | |
1997 | zb++; | |
1998 | if (zbd_zone_idx(f, zb) >= f->max_zone) | |
1999 | zb = zbd_get_zone(f, f->min_zone); | |
2000 | } while (!zb->has_wp); | |
2001 | ||
2002 | zone_lock(td, f, zb); | |
2003 | } | |
2004 | ||
a4807046 | 2005 | if (!zbd_write_zone_get(td, f, zb)) { |
4d4c71e6 | 2006 | zone_unlock(zb); |
a4807046 | 2007 | zb = zbd_convert_to_write_zone(td, io_u); |
1c74aadc | 2008 | if (!zb) { |
a4807046 | 2009 | dprint(FD_IO, "%s: can't convert to write target zone", |
1c74aadc | 2010 | f->file_name); |
59b07544 | 2011 | goto eof; |
1c74aadc | 2012 | } |
59b07544 | 2013 | } |
139d8dc6 | 2014 | |
e1a1b59b SK |
2015 | if (zbd_zone_remainder(zb) > 0 && |
2016 | zbd_zone_remainder(zb) < min_bs) | |
2017 | goto retry; | |
2018 | ||
a7c2b6fc BVA |
2019 | /* Check whether the zone reset threshold has been exceeded */ |
2020 | if (td->o.zrf.u.f) { | |
d56a6df3 SK |
2021 | if (zbdi->wp_valid_data_bytes >= |
2022 | f->io_size * td->o.zrt.u.f && | |
139d8dc6 | 2023 | zbd_dec_and_reset_write_cnt(td, f)) |
a7c2b6fc | 2024 | zb->reset_zone = 1; |
a7c2b6fc | 2025 | } |
139d8dc6 | 2026 | |
bfbdd35b BVA |
2027 | /* Reset the zone pointer if necessary */ |
2028 | if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { | |
cef8006c SK |
2029 | if (td->o.verify != VERIFY_NONE) { |
2030 | /* | |
2031 | * Unset io-u->file to tell get_next_verify() | |
2032 | * that this IO is not requeue. | |
2033 | */ | |
2034 | io_u->file = NULL; | |
2035 | if (!get_next_verify(td, io_u)) { | |
2036 | zone_unlock(zb); | |
2037 | return io_u_accept; | |
2038 | } | |
2039 | io_u->file = f; | |
2040 | } | |
2041 | ||
bfbdd35b BVA |
2042 | /* |
2043 | * Since previous write requests may have been submitted | |
2044 | * asynchronously and since we will submit the zone | |
2045 | * reset synchronously, wait until previously submitted | |
2046 | * write requests have completed before issuing a | |
2047 | * zone reset. | |
2048 | */ | |
2049 | io_u_quiesce(td); | |
2050 | zb->reset_zone = 0; | |
2051 | if (zbd_reset_zone(td, f, zb) < 0) | |
2052 | goto eof; | |
236d23a8 SK |
2053 | |
2054 | if (zb->capacity < min_bs) { | |
1c74aadc | 2055 | td_verror(td, EINVAL, "ZCAP is less min_bs"); |
07fc3f57 | 2056 | log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n", |
ee5e3436 | 2057 | zb->capacity, min_bs); |
236d23a8 SK |
2058 | goto eof; |
2059 | } | |
bfbdd35b | 2060 | } |
139d8dc6 | 2061 | |
bfbdd35b BVA |
2062 | /* Make writes occur at the write pointer */ |
2063 | assert(!zbd_zone_full(f, zb, min_bs)); | |
ee3696bd | 2064 | io_u->offset = zb->wp; |
bfbdd35b | 2065 | if (!is_valid_offset(f, io_u->offset)) { |
1c74aadc DF |
2066 | td_verror(td, EINVAL, "invalid WP value"); |
2067 | dprint(FD_ZBD, "%s: dropped request with offset %llu\n", | |
2068 | f->file_name, io_u->offset); | |
bfbdd35b BVA |
2069 | goto eof; |
2070 | } | |
139d8dc6 | 2071 | |
bfbdd35b BVA |
2072 | /* |
2073 | * Make sure that the buflen is a multiple of the minimal | |
2074 | * block size. Give up if shrinking would make the request too | |
2075 | * small. | |
2076 | */ | |
2077 | new_len = min((unsigned long long)io_u->buflen, | |
236d23a8 | 2078 | zbd_zone_capacity_end(zb) - io_u->offset); |
bfbdd35b BVA |
2079 | new_len = new_len / min_bs * min_bs; |
2080 | if (new_len == io_u->buflen) | |
2081 | goto accept; | |
2082 | if (new_len >= min_bs) { | |
2083 | io_u->buflen = new_len; | |
2084 | dprint(FD_IO, "Changed length from %u into %llu\n", | |
2085 | orig_len, io_u->buflen); | |
2086 | goto accept; | |
2087 | } | |
139d8dc6 | 2088 | |
1c74aadc | 2089 | td_verror(td, EIO, "zone remainder too small"); |
07fc3f57 | 2090 | log_err("zone remainder %lld smaller than min block size %"PRIu64"\n", |
1c74aadc | 2091 | (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); |
139d8dc6 | 2092 | |
bfbdd35b | 2093 | goto eof; |
139d8dc6 | 2094 | |
bfbdd35b | 2095 | case DDIR_TRIM: |
e3be810b SK |
2096 | /* Check random trim targets a non-empty zone */ |
2097 | if (!td_random(td) || zb->wp > zb->start) | |
2098 | goto accept; | |
2099 | ||
2100 | /* Find out a non-empty zone to trim */ | |
2101 | zone_unlock(zb); | |
39e06ee7 | 2102 | zl = zbd_get_zone(f, f->max_zone); |
e3be810b SK |
2103 | zb = zbd_find_zone(td, io_u, 1, zb, zl); |
2104 | if (zb) { | |
2105 | io_u->offset = zb->start; | |
2106 | dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n", | |
2107 | f->file_name, io_u->offset); | |
2108 | goto accept; | |
2109 | } | |
139d8dc6 | 2110 | |
e3be810b | 2111 | goto eof; |
139d8dc6 | 2112 | |
bfbdd35b | 2113 | case DDIR_SYNC: |
e3be810b | 2114 | /* fall-through */ |
bfbdd35b BVA |
2115 | case DDIR_DATASYNC: |
2116 | case DDIR_SYNC_FILE_RANGE: | |
2117 | case DDIR_WAIT: | |
2118 | case DDIR_LAST: | |
2119 | case DDIR_INVAL: | |
2120 | goto accept; | |
2121 | } | |
2122 | ||
2123 | assert(false); | |
2124 | ||
2125 | accept: | |
43bcbd5b | 2126 | assert(zb->has_wp); |
b7694961 | 2127 | assert(zb->cond != ZBD_ZONE_COND_OFFLINE); |
d9ed3e63 DLM |
2128 | assert(!io_u->zbd_queue_io); |
2129 | assert(!io_u->zbd_put_io); | |
139d8dc6 | 2130 | |
d9ed3e63 DLM |
2131 | io_u->zbd_queue_io = zbd_queue_io; |
2132 | io_u->zbd_put_io = zbd_put_io; | |
139d8dc6 | 2133 | |
2ef3c1b0 DF |
2134 | /* |
2135 | * Since we return with the zone lock still held, | |
2136 | * add an annotation to let Coverity know that it | |
2137 | * is intentional. | |
2138 | */ | |
2139 | /* coverity[missing_unlock] */ | |
139d8dc6 | 2140 | |
bfbdd35b BVA |
2141 | return io_u_accept; |
2142 | ||
2143 | eof: | |
43bcbd5b | 2144 | if (zb && zb->has_wp) |
4d4c71e6 | 2145 | zone_unlock(zb); |
139d8dc6 | 2146 | |
bfbdd35b BVA |
2147 | return io_u_eof; |
2148 | } | |
fd5d733f BVA |
2149 | |
2150 | /* Return a string with ZBD statistics */ | |
2151 | char *zbd_write_status(const struct thread_stat *ts) | |
2152 | { | |
2153 | char *res; | |
2154 | ||
ee5e3436 | 2155 | if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0) |
fd5d733f BVA |
2156 | return NULL; |
2157 | return res; | |
2158 | } | |
e3be810b SK |
2159 | |
2160 | /** | |
2161 | * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim | |
2162 | * | |
2163 | * @td: FIO thread data. | |
2164 | * @io_u: FIO I/O unit. | |
2165 | * | |
2166 | * It is assumed that z->mutex is already locked. | |
2167 | * Return io_u_completed when reset zone succeeds. Return 0 when the target zone | |
2168 | * does not have write pointer. On error, return negative errno. | |
2169 | */ | |
2170 | int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u) | |
2171 | { | |
2172 | struct fio_file *f = io_u->file; | |
2173 | struct fio_zone_info *z; | |
e3be810b SK |
2174 | int ret; |
2175 | ||
53aa6171 | 2176 | z = zbd_offset_to_zone(f, io_u->offset); |
e3be810b SK |
2177 | if (!z->has_wp) |
2178 | return 0; | |
2179 | ||
2180 | if (io_u->offset != z->start) { | |
139d8dc6 DLM |
2181 | log_err("Trim offset not at zone start (%lld)\n", |
2182 | io_u->offset); | |
e3be810b SK |
2183 | return -EINVAL; |
2184 | } | |
2185 | ||
2186 | ret = zbd_reset_zone((struct thread_data *)td, f, z); | |
2187 | if (ret < 0) | |
2188 | return ret; | |
2189 | ||
2190 | return io_u_completed; | |
2191 | } |